Methods for predicting a response to bevacizumab or platinum-based chemotherapy or both in patients with ovarian cancer

This application is the § 371 U.S. National Stage of International Application No. PCT/US2019/059218, filed Oct. 31, 2019, which claims priority to U.S. Provisional Patent Application No. 62/753,274 filed Oct. 31, 2018, each of which is incorporated herein by reference in its entirety.

This invention was made with government support under CA077598 and TR002494 awarded by the National Institutes of Health. The government has certain rights in the invention.

Epithelial ovarian cancer has the highest mortality rate of all gynecologic cancers with most patients diagnosed with stage III or IV disease. Additionally, up to one-third of patients will not respond to standard initial treatment including cytoreductive surgery and platinum-based chemotherapy. Although significant improvements in median progression-free survival (PFS) have been observed when bevacizumab was added to standard therapy, a subgroup of patients do not benefit from the treatment.

This disclosure describes methods of predicting the response of a patient with ovarian cancer to platinum-based chemotherapy and/or treatment with a monoclonal antibody against VEGF-A, bevacizumab (also referred to by the brand name AVASTIN), using clinical and molecular tumor characteristics in patients. This disclosure further provides methods of treating patients with ovarian cancer based on those predictions.

In one aspect, this disclosure describes a method for treating a patient suffering from ovarian cancer following removal of a tumor. In some embodiments, the method includes determining whether the patient is predicted to benefit from the administration of bevacizumab and, if the patient is predicted to benefit from the administration of bevacizumab, administering bevacizumab. Determining whether the patient is predicted to benefit from the administration of bevacizumab may include determining whether the patient is predicted to benefit from the administration of bevacizumab in addition to the administration of platinum-based chemotherapy.

Determining whether the patient is predicted to benefit from the administration of bevacizumab may include determining the patient's gene expression level of microfibril associated protein 2 (MFAP2) and determining the patient's gene expression level of vascular endothelial growth factor A (VEGFA). Determining whether the patient is predicted to benefit from the administration of bevacizumab may further include at least one of determining the patient's International Federation of Gynecology and Obstetrics (FIGO) stage; determining the patient's Eastern Cooperative Oncology Group (ECOG) performance status; and determining the size of the tumor tissue remaining post-removal of the tumor.

In another aspect, this disclosure describes a method for treating a patient suffering from ovarian cancer following removal of a tumor, the method comprising determining whether the patient is predicted to benefit from the administration of a platinum-based chemotherapy and, if the patient is predicted to benefit from the administration of platinum-based chemotherapy, administering platinum-based chemotherapy.

In some embodiments, determining whether the patient is predicted to respond to the administration a platinum-based chemotherapy includes determining the patient's gene expression level of microfibril associated protein 2 (MFAP2); determining the patient's International Federation of Gynecology and Obstetrics (FIGO) stage; determining the patient's Eastern Cooperative Oncology Group (ECOG) performance status; and determining the size of the tumor tissue remaining post-removal of the tumor. In some embodiments, determining whether the patient is predicted to benefit from the administration of a platinum-based chemotherapy further includes determining the patient's gene expression level of vascular endothelial growth factor A (VEGFA).

In a further aspect, this disclosure describes a method that includes identifying a patient with ovarian cancer, and determining the patient's gene expression levels of microfibril associated protein 2 (MFAP2) and vascular endothelial growth factor A (VEGFA) in a biological sample containing cancer cells obtained from the patient, determining the patient's International Federation of Gynecology and Obstetrics (FIGO) stage, determining the patient's Eastern Cooperative Oncology Group (ECOG) performance status, determining the size of the tumor tissue remaining post-removal of a tumor, and calculating a patient risk score for the patient.

In another aspect, this disclosure describes a method for predicting the response of a patient with ovarian cancer to treatment with bevacizumab. In some embodiments, the method includes: determining gene expression levels of VEGFA and MFAP2; calculating a FIGO numeric score, wherein the FIGO stage is coded as an integer; calculating a surgical outcome score, wherein the score is −1 if the surgical outcome was suboptimal; 0 if the surgical outcome was optimal but tumor tissue smaller than 1 cm remained; or +1 if the surgical outcome was optimal and no visible macroscopic tumor tissue remained; calculating an ECOG score of 0 to 2, based on ECOG performance status; and applying the expression levels, FIGO numeric score, surgical outcome score, and ECOG score to a predictive model that relates the variables with progression-free survival of ovarian cancer; and evaluating an output of the predictive model to predict progression-free survival of the patient.

In yet another aspect, this disclosure describes a method for predicting the response of a patient with ovarian cancer to treatment with bevacizumab wherein the method includes determining gene expression levels of a collection of genes taken from a biological sample of the patient, applying the expression levels to a predictive model that relates the expression levels of the collection of genes the likelihood of progression-free survival of the patient; and evaluating an output of the predictive model to predict the likelihood of progression-free survival of the patient. In some embodiments, the collection of genes comprises at least 80%, at least 90%, at least 95%, at least 98%, or 100% of the genes of any one of Tables 9-12 In some embodiments, the collection of genes comprises the genes of any one of Tables 9-12. In some embodiments, the method further includes applying at least one of FIGO stage, surgical outcome, ECOG score, and tumor histology to the predictive model.

In a further aspect, this disclosure provides a method for predicting progression-free survival of a patient with ovarian cancer. In some embodiment the method includes determining gene expression levels of a collection of genes taken from a biological sample of the patient, applying the expression levels to a predictive model that relates the expression levels of the collection of genes with progression-free survival of ovarian cancer; and evaluating an output of the predictive model to predict progression-free survival of the patient.

In some embodiments, the collection of genes includes at least 80%, at least 90%, at least 95%, at least 98%, or 100% of the genes of any one of Tables 6, 7, or 13-68. In some embodiments, the collection of genes includes the genes of any one of Tables 6, 7, or 13-68. In some embodiments, the method further includes applying at least one of FIGO stage, surgical outcome, ECOG score, and tumor histology to the predictive model.

In an additional aspect, this disclosure describes a method for predicting an outcome for a patient, the method including: receiving an identified set of biomarkers determined based on a set of predetermined data comprising clinical data, gene expression data, or both; identifying other sets of biomarkers based on the identified set of biomarkers and remaining data comprising the set of predetermined data excluding the identified set of biomarkers; generating a signature for each set of biomarkers to predict an outcome for a patient having ovarian cancer; and determining a prediction of an outcome for a patient having ovarian cancer based on one or more signatures and patient test data comprising clinical data, gene expression data, or both.

As used herein, the term “ovarian cancer” is used in the broadest sense and refers to all stages and all forms of cancer arising from the ovary.

As used herein, the term “signature” refers to a computational or mathematical model including a set of variables and corresponding coefficients. The variables may include clinical variables or molecular variables (for example, gene expression) or both. A signature may be used to evaluate patient test data.

As used herein, the term “ensemble” refers to a collection of or catalogue of signatures.

The words “preferred” and “preferably” refer to embodiments of the invention that may afford certain benefits, under certain circumstances. However, other embodiments may also be preferred, under the same or other circumstances. Furthermore, the recitation of one or more preferred embodiments does not imply that other embodiments are not useful and is not intended to exclude other embodiments from the scope of the invention.

The terms “comprises” and variations thereof do not have a limiting meaning where these terms appear in the description and claims.

Unless otherwise specified, “a,” “an,” “the,” and “at least one” are used interchangeably and mean one or more than one.

Also herein, the recitations of numerical ranges by endpoints include all numbers subsumed within that range (for example, 1 to 5 includes 1, 1.5, 2, 2.75, 3, 3.80, 4, 5, etc.).

For any method disclosed herein that includes discrete steps, the steps may be conducted in any feasible order. And, as appropriate, any combination of two or more steps may be conducted simultaneously.

The above summary of the present invention is not intended to describe each disclosed embodiment or every implementation of the present invention. The description that follows more particularly exemplifies illustrative embodiments. In several places throughout the application, guidance is provided through lists of examples, which examples can be used in various combinations. In each instance, the recited list serves only as a representative group and should not be interpreted as an exclusive list.

Reference throughout this specification to “one embodiment,” “an embodiment,” “certain embodiments,” or “some embodiments,” etc., means that a particular feature, configuration, composition, or characteristic described in connection with the embodiment is included in at least one embodiment of the disclosure. Thus, the appearances of such phrases in various places throughout this specification are not necessarily referring to the same embodiment of the disclosure. Furthermore, the particular features, configurations, compositions, or characteristics may be combined in any suitable manner in one or more embodiments.

All headings are for the convenience of the reader and should not be used to limit the meaning of the text that follows the heading, unless so specified.

Unless otherwise indicated, all numbers expressing quantities of components, molecular weights, and so forth used in the specification and claims are to be understood as being modified in all instances by the term “about.” Accordingly, unless otherwise indicated to the contrary, the numerical parameters set forth in the specification and claims are approximations that may vary depending upon the desired properties sought to be obtained by the present invention. At the very least, and not as an attempt to limit the doctrine of equivalents to the scope of the claims, each numerical parameter should at least be construed in light of the number of reported significant digits and by applying ordinary rounding techniques.

Notwithstanding that the numerical ranges and parameters setting forth the broad scope of the invention are approximations, the numerical values set forth in the specific examples are reported as precisely as possible. All numerical values, however, inherently contain a range necessarily resulting from the standard deviation found in their respective testing measurements.

FIG. 1 shows the methodological benefits (for example, computational modeling advantages) of tying development of precision medicine tests to randomized clinical trials (RCTs).

FIG. 2 shows sequential Nested N-Fold Cross-Validation model selection and error estimation design (NNFCV) used for overfitting-resistant multi-stage analysis as new methods and new data become available.

FIG. 3 shows Kaplan-Meier curves (top) and heatmaps (bottom) corresponding to subgroups and predictor variables in the reduced model identifying patients and subgroups that will benefit the most or the least from bevacizumab, as further described in Example 1.

FIG. 4 A - FIG. 4 B shows exemplary clinical strategies using precision treatment models/tests as described herein. FIG. 4 A identifies a “clear benefit” group that should receive bevacizumab; a “no benefit” group; and an intermediate group with “minor/questionable benefit” from bevacizumab. FIG. 4 B shows a strategy that combines the “no benefit” and “minor/questionable benefit” subgroups of FIG. 4 A .

In one aspect, this disclosure describes methods of determining whether a patient with ovarian cancer is predicted to benefit from platinum-based chemotherapy and/or administration of bevacizumab. The prediction may be based on the patient's clinical characteristics or molecular tumor characteristics or both. In another aspect, this disclosure provides methods of treating patients with ovarian cancer. In some embodiments, the patients may be treated based on the predictions. In another aspect, this disclosure describes a method of determining a risk score for a patient with ovarian cancer. In an additional aspect, this disclosure describes predicting progression-free survival of a patient with ovarian cancer. In a further aspect, this disclosure describes an apparatus, a system, and a kit for performing all or part of the methods described herein.

Need for and Benefit of a Predictive Test

Patients are considered platinum-refractory if they progress while on treatment or platinum-resistant if their disease recurs less than six months from completion of the initial platinum-based chemotherapy. Even in patients who have a complete initial response to chemotherapy, 80% will recur and eventually develop resistance to multiple drugs and die from drug-resistant disease. Efforts are ongoing to study novel, targeted agents, including bevacizumab, an anti-angiogenic monoclonal antibody against vascular endothelial growth factor (VEGF). Two phase III frontline trials in ovarian cancer (ICON? and GOG 218) showed statistically significant improvements in median progression-free survival (PFS) of 2.3 and 3.8 months, respectively, when bevacizumab was added to standard first-line chemotherapy (Kommoss et al. Clin Cancer Res Off J Am Assoc Cancer Res. 2017; 23(14):3794-801; Perren et al. N Engl J Med. 2011; 365(26):2484-96.) A subgroup of patients benefits significantly whereas the majority benefit moderately or do not benefit. The problem is further compounded by the high cost of bevacizumab which is currently $400,000 per progression-free life saved in the USA, thus making treatment of all patients economically infeasible. Moreover, the patients who can afford the drug are not necessarily the ones who will benefit from it. These problems underscore the pressing clinical need for more individualized treatment strategies.

At the time of the invention, gene expression analysis of ovarian cancers performed in The Cancer Genome Atlas (TCGA) had led to a molecular classification of ovarian cancer into four subtypes (Tothill et al. Clin Cancer Res Off J Am Assoc Cancer Res. 2008; 14(16):5198-208; Konecny et al. J Natl Cancer Inst. 2014; 106(10):dju249; Winterhoff et al. Gynecol Oncol. 2016; 141(1):95-100.) These four subgroups have some prognostic significance. (Winterhoff et al. Gynecol Oncol. 2016; 141(1):95-100; Konecny et al. J Natl Cancer Inst. 2014; 106(10):dju249.) Although differential response to bevacizumab and platinum-based chemotherapy within those four molecular subtypes had been demonstrated using formalin-fixed paraffin-embedded (FFPE) tumor samples (Kommoss et al. Clin Cancer Res Off J Am Assoc Cancer Res. 2017; 23(14):3794-801), development and statistical validation of a clinico-molecular stratification model with sufficient accuracy was needed to allow these observations to be clinically actionable. Development of such a model is described in the present disclosure (Example 1).

The potential for health economic impact of a precision test based on the predictivity of the models and corresponding clinical strategies described herein is enormous. For example, if only patients who were predicted to strongly benefit from treatment with bevacizumab were treated instead of all patients, up to $90 billion in savings globally could be realized over 10 years. Moreover, the methods described herein may identify patients who will not benefit from either conventional or bevacizumab treatment, allowing them to be routed to alternative experimental treatments, providing additional survival and economic benefits.

Determining Gene Expression Levels

In some embodiments, the methods described herein include determining a gene expression level of a patient.

In some embodiments, a gene expression level may be measured using a standard biochemical technique and/or assay and may be converted to a quantitative gene expression level using an appropriate value transformation for that technology. In some embodiments, the gene expression level may be used as an input in a model, as described herein.

In some embodiments, the gene includes microfibril associated protein 2 (MFAP2) or vascular endothelial growth factor A (VEGFA) or both.

In some embodiments, determining a gene expression level of a patient includes determining the gene expression level of a collection of genes taken from a biological sample of the patient.

In some embodiments, the collection of genes includes the genes of any one of Tables 6, 7, or 9-68. In some embodiments, some of the genes of a table may be excluded from the collection of genes at the cost of some reduction in predictive performance. In some embodiments, the collection of genes includes at least two genes, at least 14 genes, at least 18 genes, at least 20 genes, or at least 30 genes selected from the genes of any one of Tables 6, 7, or 9-68. In some embodiments, the collection of genes includes at least 50%, at least 60%, at least 70%, at least 80%, at least 90%, at least 95%, at least 98%, or 100% of the genes of any one of Tables 6, 7, or 9-68.

In some embodiments, the collection of genes of any one of Tables 6, 7, or 9-68 may be selected by excluding only those genes of that table that do not significantly affect predictivity.

In some embodiments, the collection of genes may be selected by optimizing predictivity with a constraint or a set of constraints. A constraint may include, for example, cost or user convenience.

In some embodiments, determining a gene expression level includes assessing the amount (for example, absolute amount, relative amount, or concentration) of a gene product in a sample. A gene product may include, for example, a protein or RNA transcript encoded by the gene, or a fragment of the protein or RNA transcript. In some embodiments, determining a gene expression level includes receiving the results of such an assessment. In some embodiments, determining a gene expression level includes converting the results of such an assessment to a quantitative gene expression level.

A sample may include a biological sample of the patient. In some embodiments, the sample may be a biological sample containing cancer cells. For example, the sample may include a tissue sample obtained by biopsy of a patient, a bodily fluid (for example, blood, plasma, serum, urine, etc.), a cell that is the progeny of a patient's tumor cell, or a sample enriched for tumor cells.

The sample may be subjected to a variety of well-known post-collection preparative and storage techniques (for example, fixation, storage, freezing, lysis, homogenization, DNA or RNA extraction, ultrafiltration, concentration, evaporation, centrifugation, etc.) prior to assessing the gene expression level in the sample.

The amount of the gene product may be assessed by any suitable method known to a person having skill in the art. For example, gene expression may be identified using sequencing, quantitative RT-PCR, microarray analysis, and/or immunohistochemistry as described in, for example, U.S. Pat. No. 8,725,426 and WO 2015/109234. Standard assay normalization methods and batch effect correction methods suitable to each type of assay may also be employed.

In some embodiments, the method includes normalizing the gene expression levels including, for example, normalizing the level of the RNA transcripts to obtain normalized gene expression levels.

International Federation of Gynecology and Obstetrics (FIGO) Stage

In some embodiments, the methods of this disclosure include determining a patient's International Federation of Gynecology and Obstetrics (FIGO) stage, as described at, for example, www.cancer.org/cancer/ovarian-cancer/detection-diagnosis-staging/staging.html. In some embodiments, the FIGO stage may be coded as an integer for the purposes of calculating a risk score for a patient. For example, FIGO stage IA=1, FIGO stage IB=2, FIGO stage IC=3, FIGO stage IIA=4, FIGO stage IIB=5, FIGO stage IIC=5, FIGO stage IIIA=7, FIGO stage IIIB=8, FIGO stage IIIC=9, and FIGO stage IV=10.

Eastern Cooperative Oncology Group (ECOG) Performance Status

In some embodiments, the methods of this disclosure include determining a patient's Eastern Cooperative Oncology Group (ECOG) performance status. Oken et al. Am J Clin Oncol. 1982; 5:649-655. A patient has an ECOG performance status of 0 if the patient is fully active and able to carry on all pre-disease performance without restriction. A patient has an ECOG performance status of 1 if the patient is restricted in physically strenuous activity but ambulatory and able to carry out work of a light or sedentary nature including, for example, light house work, office work, etc. A patient has an ECOG performance status of 2 if the patient is ambulatory and capable of all selfcare but unable to carry out any work activities and is up and about more than 50% of waking hours. A patient has an ECOG performance status of 3 if the patient is capable of only limited selfcare; confined to bed or chair more than 50% of waking hours. A patient has an ECOG performance status of 4 if the patient is completely disabled.

Removal of the Tumor and Size of the Tumor Tissue

In some embodiments, the methods of this disclosure include treating a patient after removal of a tumor by surgery. In some embodiments, the methods of this disclosure include determining the size of a patient's tumor after removal of the tumor.

Removal of the ovarian cancer (including, for example, the tumor) by surgery may include any surgical method undertaken for the removal of cancerous surgery including, for example, hysterectomy, oophorectomy, salpingo-oophorectomy, omentectomy, and/or removal of any visible cancer within the abdomen including, for example, resection of bowel, parts of the liver spleen, a lymph node, diaphragm, parts of the stomach and or pancreas, gallbladder, and any other involved tissue or organ.

In some embodiments, the tumor may be a primary tumor (for example, from the ovary, fallopian tube or primary peritoneum). In some embodiments, the tumor may be a secondary tumor (for example, a metastatic tumor from a different organ to the ovary and or fallopian tube).

In some embodiments, a patient may be characterized based on whether the surgical outcome was suboptimal (that is, tumor tissue greater than 1 centimeter (cm) remained); the surgical outcome was optimal (that is, no tumor tissue greater than 1 cm remained) but tumor tissue smaller than 1 cm remained; or the surgical outcome was optimal and no visible macroscopic tumor tissue remained. In some embodiments, the patient's surgical outcome may be converted to a score (surg_outcome), where surg_outcome is −1 if the surgical outcome was suboptimal; surg_outcome is 0 if the surgical outcome was optimal but tumor tissue smaller than 1 cm remained; and surg_outcome is +1 if the surgical outcome was optimal and no visible macroscopic tumor tissue remained.

Tumor Histology

In some embodiments, a patient may be characterized based on the histology of the tumor as determined by a pathologist. For example, microscopic examination of tumor tissue by a pathologist may be used to determine whether a patient has a serous borderline ovarian tumor (hist_rev_SBOT) or a metastatic tumor (hist_rev_metastais). If the patient is found to have a tumor (for example, either a serous borderline ovarian tumor or a metastatic tumor), the patient may be assigned a value: 1; if a tumor is present, 0 if a tumor is not present.

Platinum-Based Chemotherapy and Administration of Platinum-Based Chemotherapy

In some embodiments, the methods described herein include determining whether a patient is predicted to benefit from the administration of platinum-based chemotherapy. In some embodiments, the methods described herein include administering platinum-based chemotherapy. In some embodiments, the methods described herein include administering platinum-based chemotherapy if a patient is predicted to benefit from the administration of platinum-based chemotherapy. In some embodiments, the methods described herein include administering platinum-based chemotherapy in combination with bevacizumab.

Platinum-based chemotherapy may include any suitable platinum-based chemotherapy. Platinum-based chemotherapy may include, for example, one or more of cisplatin, carboplatin, oxaliplatin, nedaplatin, lobaplatin, heptaplatin, triplatin tetranitrate, phenanthriplatin, picoplatin, and satraplatin.

Platinum-based chemotherapy may be administered by any suitable method. The selected dosage level will depend upon a variety of factors including the activity of the particular compound of the present disclosure employed, the route of administration, the time of administration, the rate of excretion of the particular compound being employed, the duration of the treatment, other drugs, compounds and/or materials used in combination with the chemotherapy, the age, sex, weight, condition, general health and prior medical history of the subject being treated, and like factors well known in the medical arts.

Bevacizumab and Administration of Bevacizumab

In some embodiments, the methods described herein include determining whether a patient is predicted to benefit from the administration of bevacizumab. In some embodiments, the methods described herein include administering bevacizumab. In some embodiments, the methods described herein include administering bevacizumab if a patient is predicted to benefit from the administration of bevacizumab. In some embodiments, the methods described herein include administering bevacizumab in combination with platinum-based chemotherapy. In some embodiments, the methods described herein include administering bevacizumab in combination with platinum-based chemotherapy if a patient is predicted to benefit from the administration of bevacizumab in combination with platinum-based chemotherapy.

In some embodiments, determining whether a patient is predicted to benefit from the administration of bevacizumab may include using one or more of the sets of variables enumerated in Tables 9-12. In some embodiments, a set of variables (that is the set of genes and other biomarkers) as enumerated in one of Tables 9-12 may be used in combination with the corresponding coefficients described in those tables. In some embodiments, a set of variables (as enumerated in one of Tables 9-12 may be used in combination with alternative coefficients including, for example, coefficients obtained using a fitting protocol and classifier as described herein.

In some embodiments, determining whether a patient is predicted to benefit from the administration of bevacizumab may include using one or more of the sets of genes enumerated in Tables 9-12. In some embodiments, a set of gene as enumerated in one of Tables 9-12 may be used in combination with the corresponding coefficients described in those tables. In some embodiments, a set of genes of one of Tables 9-12 may be used in combination with alternative coefficients including, for example, coefficients obtained using a fitting protocol and classifier as described herein.

In some embodiments, determining whether a patient is predicted to benefit from the administration of bevacizumab may include using one or more of the sets of genes enumerated in Tables 9-12.

Bevacizumab may be administered by any suitable method. The selected dosage level will depend upon a variety of factors including the activity of the particular compound employed, the route of administration, the time of administration, the rate of excretion of bevacizumab, the rate of metabolism of bevacizumab, the duration of the treatment, other drugs, compounds and/or materials used in combination with bevacizumab, the age, sex, weight, condition, general health and prior medical history of the subject being treated, and like factors well known in the medical arts.

Predictive Model

In some embodiments, a method described herein includes determining if a patient is predicted to benefit from the administration of bevacizumab, including the administration of bevacizumab in combination with platinum-based chemotherapy. In some embodiments, a method described herein includes determining if a patient is predicted to benefit from the administration of a platinum-based chemotherapy (for example, a platinum-based chemotherapy with bevacizumab or a platinum-based chemotherapy without bevacizumab). In some embodiments, the method may include determining if a patient is predicted to benefit from the administration of bevacizumab in combination with the administration of platinum-based chemotherapy. In some embodiments, the method may include predicting progression-free survival of the patient or the difference in progression-free survival of the patient depending on which therapy is administered.

In some embodiments, the method includes determining the patient's gene expression level of microfibril associated protein 2 (MFAP2); and/or determining the patient's gene expression level of vascular endothelial growth factor A (VEGFA). In some embodiments, the method may further include one or more of: determining the patient's International Federation of Gynecology and Obstetrics (FIGO) stage; determining the patient's Eastern Cooperative Oncology Group (ECOG) performance status; and determining the size of the tumor tissue remaining post-removal of the tumor.

In some embodiments, a threshold gene expression level of MFAP may be selected based on a clinical outcome (for example, a certain increase in progression free survival), and an expression level greater than that threshold expression may indicate an increased likelihood of benefit from the administration of bevacizumab, In some embodiments, a threshold gene expression level of VEGFA may be selected based on a clinical outcome (for example, a certain increase in progression free survival) and a gene expression level greater than that threshold expression may indicate a decreased likelihood of benefit from the administration of bevacizumab. In some embodiments, a FIGO stage greater than 1 may indicate a decreased likelihood of benefit from the administration of bevacizumab. In some embodiments, an ECOG performance status greater than 0 may indicate an increased likelihood of benefit from the administration of bevacizumab. In some embodiments, a tumor size smaller than 1 cm may indicate an increased likelihood of benefit from the administration of bevacizumab. In some embodiments, a threshold value of the combinations of the MFAP, VEGFA, FIGO stage and ECOG values may be selected based on a clinical outcome (for example, a certain increase in progression free survival) and a value of the combination greater than that threshold expression may indicate a decreased likelihood of benefit from the administration of bevacizumab.

In some embodiments, a threshold gene expression level of MFAP may by selected based on a clinical outcome (for example, a certain increase in progression free survival), and a gene expression level greater than that threshold gene expression level may indicate a decreased likelihood of benefit from the administration of platinum-based chemotherapy. In some embodiments, a threshold gene expression level of VEGFA may by selected based on a clinical outcome (for example, a certain increase in progression free survival) and an expression level greater than that threshold gene expression level may indicate an increased likelihood of benefit from the administration of platinum-based chemotherapy. In some embodiments, a FIGO stage greater than 1 may indicate a decreased likelihood of benefit from the administration of platinum-based chemotherapy. In some embodiments, an ECOG performance status greater than 0 may indicate a decreased likelihood of benefit from the administration of platinum-based chemotherapy. In some embodiments, a tumor size smaller than 1 cm may indicate an increased likelihood of benefit from the administration of platinum-based chemotherapy. In some embodiments, a threshold value of the combinations of the MFAP, VEGFA, FIGO stage and ECOG values may be selected based on a clinical outcome (for example, a certain increase in progression free survival) and a value of the combination greater than that threshold expression may indicate an increased likelihood of benefit from the administration of platinum-based chemotherapy.

In some embodiments, the method may include determining a patient's predicted progression-free survival. For example, the method may include determining if a patient's predicted progression-free survival time with the administration of a platinum-based chemotherapy and bevacizumab and/or determining the patient's predicted progression-free survival time with the administration of a platinum-based chemotherapy without bevacizumab. In some embodiments, the method may include comparing the patient's predicted progression-free survival time with the administration of a platinum-based chemotherapy and bevacizumab and the patient's predicted progression-free survival time with the administration of a platinum-based chemotherapy without bevacizumab.

In some embodiments, determining a patient's predicted progression-free survival may include using one or more of the sets of variables enumerated in Table 6, Table 7, or one or more of the sets of variables described in Example 6 (Tables 13-68). In some embodiments, a set of variables (that is the set of genes and other biomarkers) as enumerated in one of Tables 6, 7, or 13-68 may be used in combination with the corresponding coefficients described in those tables. In some embodiments, a set of variables (as enumerated in one of Tables 6, 7, or 13-68 may be used in combination with alternative coefficients including, for example, coefficients obtained using a fitting protocol and classifier as described herein.

In some embodiments, determining a patient's predicted progression-free survival may include using one or more of the sets of variables enumerated in Table 6, Table 7, or one or more of the sets of variables described in Example 6 (Tables 13-68). In some embodiments, a set of variables (that is the set of genes and other biomarkers) as enumerated in one of Tables 6, 7, or 13-68 may be used in combination with the corresponding coefficients described in those tables. In some embodiments, a set of variables (as enumerated in one of Tables 6, 7, or 13-68 may be used in combination with alternative coefficients including, for example, coefficients obtained using a fitting protocol and classifier as described herein.

In some embodiments, determining a patient's predicted progression-free survival may include using one or more of the sets of genes enumerated in Table 6, Table 7, or one or more of the sets of genes described in Example 6 (Tables 13-68). In some embodiments, a set of gene as enumerated in one of Tables 6, 7, or 13-68 may be used in combination with the corresponding coefficients described in those tables. In some embodiments, a set of genes of one of Tables 6, 7, or 13-68 may be used in combination with alternative coefficients including, for example, coefficients obtained using a fitting protocol and classifier as described herein.

In some embodiments, the method may include determining whether a patient's predicted increase in progression-free survival time with the administration of a platinum-based chemotherapy and bevacizumab compared to the patient's predicted progression-free survival time with the administration of a platinum-based chemotherapy without bevacizumab is clinically meaningful. In some embodiments, a “clinically meaningful” increase in progression-free survival time may be determined by the treating physician. In some embodiments, the method may include defining a benefit threshold.

In some embodiments, a patient may be predicted to benefit from the administration of bevacizumab if the patient's predicted increase in progression-free survival is at least 3 months, at least 4 months, at least 5 months, at least 6 months, at least 7 months, at least 8 months, at least 9 months, or at least 10 months.

In some embodiments, the method may include applying a model for modeling time-to-event outcomes (for example, progression-free survival). Any model suitable for modeling time-to-event outcomes may be used including, for example, a Cox model or an accelerated failure time model. In some embodiments, the method may include applying a model for modeling binary outcomes (for example, progression-free survival up to a certain time point). Any modeling procedure suitable for modeling binary outcomes may be used including, for example, a support vector machine model or another classification method appropriate for biomedical data classification. In some embodiments, the method may include applying Nested N-Fold Cross-Validation (NNFCV).

In some embodiments, the method may include calculating a patient risk score. For example, in some embodiments, a patient's risk score may be calculated as described in Example 3.

In some embodiments, the method may further include calculating a patient's risk of recurrence at time t. For example, in some embodiments, a patient's risk of recurrence at time t may be calculated as described in Example 3.

In some embodiments, a method may include applying a patient's gene expression level (or levels) to a predictive model that relates the expression level (or levels) with progression-free survival of ovarian cancer. In some embodiments, a method may include applying the expression levels of a collection of genes to a predictive model that relates the expression levels of that collection of genes with progression-free survival of ovarian cancer. Examples of such collections of genes are provided herein. In some embodiments, the method may further include determining, applying, or determining and applying one or more of: the patient's International Federation of Gynecology and Obstetrics (FIGO) stage; the patient's Eastern Cooperative Oncology Group (ECOG) performance status; the size of the tumor tissue remaining post-removal of the tumor; tumor histology indicating a serous borderline ovarian tumor (hist_rev_SBOT); and tumor histology indicating a metastatic tumor (hist_rev_metastasis).

In some embodiments, the method includes determining the expression level of a gene or a collection of genes multiple times.

In some embodiments, the method includes detecting an additional biomarker of progression-free survival of the patient. Such biomarkers may include, for example, a germline mutation, a somatic mutation, a DNA methylation marker, and/or a protein marker.

Predictive Ensemble Model

In some embodiments, methods for predicting an outcome for a patient include receiving an identified set of biomarkers determined based on a set of predetermined data including clinical data, gene expression data, or both; identifying other sets of biomarkers based on the identified set of biomarkers and remaining data includes the set of predetermined data excluding the identified set of biomarkers; generating a signature for each set of biomarkers to predict an outcome for a patient having ovarian cancer; and determining a prediction of an outcome for a patient having ovarian cancer based on one or more of the signatures and patient test data including clinical data, gene expression data, or both.

The identified set of biomarkers may be determined to have optimal predictivity. The identified set of biomarkers may also be determined to have non-redundancy and may be described as a “Markov Boundary” biomarker set.

In some embodiments, the outcome relates to progression-free survival for a patient with ovarian cancer. In other embodiments, the outcome relates to benefitting from the administration of bevacizumab, platinum-based chemotherapy, or both for a patient with ovarian cancer.

Any suitable identified set of biomarkers may be used. In some embodiments, the identified set of biomarkers is a member of an ensemble, which is described herein in more detail. In some embodiments, the signatures of the ensemble include some or all genes of any one of Table 6, Table 7, and Tables 9-68.

A TIE* algorithm (or other multiplicity discovery technique) may be used to identify the remaining Markov Boundary sets of biomarkers in the data other than the previously identified set of biomarkers. In some embodiments, identifying other sets of biomarkers includes feeding the previously identified set of biomarkers and remaining data into a TIE* algorithm to provide the other equivalent sets of biomarkers. In particular, the TIE* algorithm may provide equivalent sets of biomarkers to the previously identified set of biomarkers. Any other appropriate biomarker and signature multiplicity discovery technique may be used in place of the TIE* algorithm known to one skilled in the art having the benefit of this disclosure.

Any suitable instantiation of the TIE* algorithm (or algorithms with similar functionality) may be used. (Statnikov and Aliferis. PLoS Computational Biology 2010; 6(5), p. e1000790; U.S. Pat. No. 8,805,761; Aliferis et al. Journal of Machine Learning Research 2010; 11(January), pp. 171-234; Statnikov et al. Journal of Machine Learning Research 2013; 14(February), pp. 499-566; U.S. Pat. No. 8,655,821.)

In some embodiments, the TIE* algorithm systematically examines information equivalences in the “seed” biomarker set (and by extension to all corresponding optimal signatures) with variables in the remainder of the data (for example, full set of variables minus the seed). Replacement of a subset of the “seed” and execution of a subroutine may be performed to identify the Markov Boundary set of biomarkers in the remainder of the data (for example, running the subroutine once for each time a subset of the “seed” is excluded). The replacement of the subset of the “seed” and execution of the subroutine may be repeated recursively until all existing sets of biomarkers have been identified and output by the TIE* algorithm. As the TIE* algorithm, traverses the space of replacement subsets the remainder of the data shrinks. In some embodiments, the TIE* algorithm will terminate when no biomarker replacement can generate new equivalent biomarker sets.

In some embodiments, generating a signature for each set of biomarkers sets identified by TIE* (or other multiplicity algorithm) includes feeding each set of biomarkers into a machine learning classifier fitting and model “pipeline”. The pipeline may incorporate model selection and error estimation. The pipeline may apply one or more of the following: a repeated nested n-fold cross validation with grid parameter choice, a support vector machine classifier, a random forest classifier, a lasso classifier, or any other suitable technique in the field of “omics” based classification by molecular signature construction. In some embodiments, the output of the TIE* algorithm provides a catalogue, or database, of biomarker sets.

Each set of biomarkers may be fed into a machine learning classifier fitting and model pipeline that typically incorporates model selection and error estimation. (Statnikov. A gentle introduction to support vector machines in biomedicine: Theory and methods; Vol. 1. World Scientific Pub. Co.; 2011; Statnikov et al. A Gentle Introduction to Support Vector Machines in Biomedicine: Volume 2: Case Studies and Benchmarks. World Scientific Pub. Co.; 2013.) One or more methods for deriving signatures, or models, from datasets may be used. In some embodiments, different models may be generated by the pipeline. In some embodiments, different models can be generated by a machine learning classifier fitting and model pipeline. In some embodiments, different models can have the same underlying sets of biomarkers but with different coefficients for each biomarker in the set. For example, a plurality of classifier models can be produced for each set of biomarkers, each having different coefficients. Although the models may have different coefficients, the models can be constructed so that they will have functional (input-output) equivalency. Further, coefficients in each model may be refit as new data is acquired.

Still further, coefficients may be tuned to a particular measuring platform used to generate the biomarker data, such as clinical or gene expression data. Different measuring platforms may require slightly different coefficients.

The output of the pipeline for each set of biomarkers, or each member of the equivalency catalogue, may be used as a signature for predicting patient outcomes, for example, in response to treatment. Typically, a signature does not include data used for its construction or validation. These signatures may be implemented as a companion test, or companion diagnostic, according to usual methods that combine: assaying of the biomarkers from tumor tissue specimens and processing of the generated measurements via fitting and application of classifiers to create clinical decision guidance and support that is delivered in clinical practice.

In some embodiments, the signatures are statistically indistinguishable from one another for a particular predictivity level. In some embodiments, each signature is a minimal (for example, non-reducible without degradation of predictivity) set of biomarkers for a particular predictivity level.

The catalogue of signatures may be described as an ensemble. In some embodiments, determining a prediction of an outcome for a patient having ovarian cancer is based on an ensemble prediction using a plurality of the signatures. The catalogue of signatures may be used to provide an ensemble prediction. Use of the ensemble prediction may reduce, or even minimize, the variance of prediction accuracy when compared to using single signatures.

In one example, the ensemble prediction may average outputs of each of the signatures. A prediction may be obtained from every signature in the catalogue, and the predictions may be averaged to obtain a consolidated ensemble prediction.

In another example, the ensemble prediction may use a plurality of the signatures based on available patient test data. A prediction may be obtained from only a select number of signatures in the catalogue, or ensemble, and the predictions may be averaged to obtain a consolidated ensemble prediction. The signatures may be selected based on availability. In some embodiments, one or a few signatures (for example, up to the full ensemble) may be used for prediction. Factors contributing to availability, or choice of signature to use, may include one or more of: convenience, cost, and ease of collection. The companion test may be personalized or customized for different patients by means of choice of members of the ensemble of signatures.

Testing Whether a Signature Belongs in the Ensemble

A signature may be tested by a party who does not have a full ensemble to determine whether the signature belongs in an existing ensemble used to predict a particular outcome. In one example, when the full ensemble of signatures is known the inventor simply needs perform a table lookup for the signature against the ensemble. When the ensemble is not disclosed a method may determine whether the signature belongs to the existing ensemble even if all the signatures in the ensemble are unknown to the party. In general, determining the full ensemble (for example, determining all the equivalent sets of biomarkers.

The method may include determining whether the predictivity level of a signature under consideration is statistically indistinguishable from the known predictivity of the existing set of signatures in the ensemble. Any suitable statistical technique for testing differences of predictivity measures of classifiers may be used to compare the predictivity levels to determine whether the predictivity levels are statistically indistinguishable as known to one skilled having the benefit of this disclosure. (Statnikov et al. A Gentle Introduction to Support Vector Machines in Biomedicine: Volume 2: Case Studies and Benchmarks. World Scientific Pub. Co.; 2013.)

The method may also include determining whether new signature is minimal for the related predictivity level. Minimality of the new signature may be established by testing and verifying that removal of at least one subset of markers does not leave the predictivity level intact.

If the signature has a predictivity level that is statistically indistinguishable from the predictivity of signatures in the existing ensemble and the signature is minimal, then the signature may be determined to belong in the existing ensemble.

If the new signature has a predictivity level that is statistically distinguishable, then the signature is not part of the ensemble.

If the new signature has a predictivity level that is statistically indistinguishable from a known signature in the existing ensemble but is not minimal, then the method may determine that the signature includes a signature that is part of the existing ensemble (whether known or unknown) plus some noise, or redundant markers. Noise or redundant markers may be described as adding no predictive value to the signature of the ensemble.

In general, the addition of biomarkers beyond the minimal level required for optimal predictivity should not confer any predictive advantage and thus would not constitute an enhanced or otherwise improved signature. Therefore, any predictively optimal biomarker set and signature that is minimal also corresponds to a large number of biomarker sets and signatures that may be constructed by “padding” essential biomarkers with predictively unnecessary (and potentially costly and cumbersome) biomarkers.

Apparatus and Systems

The present disclosure further provides exemplary apparatuses and systems for executing all or part of the methods described herein. In some embodiments, an apparatus may include, for example, a computer, a processor, or a group of processors. In some embodiments, an apparatus may include a microarray, a sequencer, and/or a device capable of performing PCR. A system may include, for example, a computer program, a computer-readable medium, or an algorithm.

Kits

In another aspect, this disclosure describes a kit that may be used to perform all or part of a method described herein. For example, in some embodiments, a kit may include reagent suitable for determining gene expression levels. In some embodiments, a kit may include a system for executing a computer program described herein.

Exemplary Method Embodiments Including Administration of Bevacizumab

1. A method for treating a patient suffering from ovarian cancer following removal of a tumor, the method comprising:

determining whether the patient is predicted to benefit from the administration of bevacizumab, wherein such determination comprises:

determining the patient's International Federation of Gynecology and Obstetrics (FIGO) stage; determining the patient's Eastern Cooperative Oncology Group (ECOG) performance status; and determining the size of the tumor tissue remaining post-removal of the tumor.

4. The method of Embodiment 3, wherein

a gene expression level of MFAP greater than a threshold gene expression level indicates a decreased likelihood of benefit from platinum-based chemotherapy, wherein the threshold gene expression level is selected based on a clinical outcome; a gene expression level of VEGFA greater than a threshold gene expression level indicates an increased likelihood of benefit from the administration of platinum-based chemotherapy, wherein the threshold gene expression level is selected based on a clinical outcome; a FIGO stage greater than 1 indicates a decreased likelihood of benefit from the administration of bevacizumab, an ECOG performance status greater than 0 indicates an increased likelihood of benefit from the administration of bevacizumab, and a tumor size smaller than 1 cm indicates an increased likelihood of benefit from the administration of bevacizumab.

5. The method of Embodiment 4, wherein the clinical outcome comprises increased time of progression-free survival.

6. The method of Embodiment 5, wherein the patient's predicted increase in progression-free survival is at least 3 months, at least 4 months, at least 5 months, at least 6 months, at least 7 months, at least 8 months, at least 9 months, or at least 10 months.

7. The method of any one of the preceding Embodiments, wherein determining whether the patient is predicted to benefit from the administration of bevacizumab further comprises determining the patient's predicted progression-free survival time with the administration of a platinum-based chemotherapy without bevacizumab.

8. The method of Embodiment 7, wherein determining whether the patient is predicted to benefit from a platinum-based chemotherapy without bevacizumab comprises:

determining the patient's gene expression level of microfibril associated protein 2 (MFAP2); determining the patient's gene expression level of vascular endothelial growth factor A (VEGFA); determining the patient's International Federation of Gynecology and Obstetrics (FIGO) stage; determining the patient's Eastern Cooperative Oncology Group (ECOG) performance status; and determining the size of the tumor tissue remaining post-removal of the tumor.

9. The method of Embodiment 8, wherein

a gene expression level of MFAP greater than a threshold gene expression level indicates a decreased likelihood of benefit from platinum-based chemotherapy, wherein the threshold gene expression level is selected based on a clinical outcome; a gene expression level of VEGFA greater than a threshold gene expression level indicates an increased likelihood of benefit from the administration of platinum-based chemotherapy, wherein the threshold gene expression level is selected based on a clinical outcome; a FIGO stage greater than 1 indicates a decreased likelihood of benefit from platinum-based chemotherapy, an ECOG performance status greater than 0 indicates a decreased likelihood of benefit from platinum-based chemotherapy, and a tumor size smaller than 1 cm indicates an increased likelihood of benefit from platinum-based chemotherapy.

10. The method of Embodiment 9, wherein the clinical outcome comprises increased time of progression-free survival.

11. The method of Embodiment 10, wherein the patient's predicted increase in progression-free survival is at least 3 months, at least 4 months, at least 5 months, at least 6 months, at least 7 months, at least 8 months, at least 9 months, or at least 10 months.

12. The method of any one of Embodiments 7 to 11, wherein determining whether the patient is predicted to benefit from the administration of bevacizumab further comprises determining if the patient's predicted progression-free survival time with the administration of a platinum-based chemotherapy and bevacizumab is greater than the patient's predicted progression-free survival time with the administration of a platinum-based chemotherapy without bevacizumab.

13. The method of Embodiment 12, wherein the patient is predicted to benefit from the administration of bevacizumab if the patient's predicted increase in progression-free survival is clinically meaningful.

14. The method of Embodiment 13, wherein the patient is predicted to benefit from the administration of bevacizumab if the patient's predicted increase in progression-free survival is at least 3 months, at least 4 months, at least 5 months, at least 6 months, at least 7 months, at least 8 months, at least 9 months, or at least 10 months.

15. The method of any one of the preceding Embodiments, wherein determining whether the patient is predicted to benefit from the administration of bevacizumab comprises defining a benefit threshold.

16. The method of any one of the preceding Embodiments, wherein determining whether the patient is predicted to benefit from the administration of bevacizumab comprises applying a Cox model.

17. The method of any one of the preceding Embodiments, wherein the method comprises administering platinum-based chemotherapy.

18. The method of any one of the preceding Embodiments, wherein the tumor comprises a primary tumor.

19. The method of any one of the preceding Embodiments, wherein the tumor comprises a secondary tumor.

18. The method of any one of the preceding Embodiments, wherein the tumor comprises a primary tumor or a secondary tumor.

20. The method of any one of the preceding Embodiments, further comprising:

receiving an identified set of biomarkers determined based on a set of predetermined data comprising clinical data, gene expression data, or both, wherein the identified set of biomarkers comprises at least MFAP2 and VEGFA; identifying other sets of biomarkers based on the identified set of biomarkers and remaining data comprising the set of predetermined data excluding the identified set of biomarkers; and generating a signature for each set of biomarkers to predict an outcome for a patient having ovarian cancer, wherein determining whether the patient is predicted to benefit from the administration of bevacizumab is based on an ensemble prediction using a plurality of signatures and patient test data comprising clinical data, gene expression data, or both.

Exemplary Method Embodiments Including Administration of a Platinum-Based Chemotherapy

1. A method for treating a patient suffering from ovarian cancer following removal of a tumor, the method comprising:

determining whether the patient is predicted to benefit from the administration of a platinum-based chemotherapy, wherein such determination comprises:

determining the patient's gene expression level of microfibril associated protein 2 (MFAP2); determining the patient's International Federation of Gynecology and Obstetrics (FIGO) stage; determining the patient's Eastern Cooperative Oncology Group (ECOG) performance status; and determining the size of the tumor tissue remaining post-removal of the tumor; and

if the patient is predicted to benefit from the administration of platinum-based chemotherapy, administering platinum-based chemotherapy.

2. The method of Embodiment 1, wherein determining whether the patient is predicted to benefit from the administration of a platinum-based chemotherapy further comprises:

determining the patient's gene expression level of vascular endothelial growth factor A (VEGFA).

3. The method of Embodiment 2, wherein

a gene expression level of MFAP greater than a threshold gene expression level indicates a decreased likelihood of benefit from the administration of platinum-based chemotherapy, wherein the threshold gene expression level is selected based on a clinical outcome; a gene expression level of VEGFA greater than a threshold gene expression level indicates an increased likelihood of benefit from the administration of platinum-based chemotherapy, wherein the threshold gene expression level is selected based on a clinical outcome; a FIGO stage greater than 1 indicates a decreased likelihood of benefit from platinum-based chemotherapy, an ECOG performance status greater than 0 indicates aa decreased likelihood of benefit from platinum-based chemotherapy, and a tumor size smaller than 1 cm indicates an increased likelihood of benefit from platinum-based chemotherapy.

4. The method of Embodiment 3, wherein the clinical outcome comprises increased time of progression-free survival.

5. The method of Embodiment 4, wherein the patient's predicted increase in progression-free survival is at least 3 months, at least 4 months, at least 5 months, at least 6 months, at least 7 months, at least 8 months, at least 9 months, or at least 10 months.

6. The method of any one of the preceding Embodiments, wherein determining whether the patient is predicted to benefit from the administration of a platinum-based chemotherapy further comprises determining the patient's predicted progression-free survival time.

7. The method of any one of the preceding Embodiments, wherein determining whether the patient is predicted to benefit from the administration of a platinum-based chemotherapy comprises applying a Cox model.

8. The method of any one of the preceding Embodiments, wherein the method comprises administering bevacizumab.

9. The method of any one of the preceding Embodiments, wherein the tumor is a primary tumor.

Exemplary Method Embodiments Including Calculating a Quantitative Score

1. A method comprising:

identifying a patient with ovarian cancer; determining a patient's gene expression levels of microfibril associated protein 2 (MFAP2) and vascular endothelial growth factor A (VEGFA) in a biological sample containing cancer cells obtained from the patient, determining the patient's International Federation of Gynecology and Obstetrics (FIGO) stage, determining the patient's Eastern Cooperative Oncology Group (ECOG) performance status, determining the size of the tumor tissue remaining post-removal of a tumor, and calculating a patient risk score for the patient.

2. The method of Embodiment 1, wherein the patient risk score (recurrence_score) is calculated as follows:

recurrence_score=0.31*figo_numeric−0.35*surg_outcome+0.23*MFAP2+0.48*ECOG+0.19*VEGFA*Bevacizumab−0.15*MFAP2*Bevacizumab−0.44*ECOG*Bevacizumab

wherein figo_numeric=FIGO stage coded as integers,

wherein surg_outcome is −1 if the surgical outcome was suboptimal; 0 if the surgical outcome was optimal but tumor tissue smaller than 1 cm remained; or +1 if the surgical outcome was optimal and no visible macroscopic tumor tissue remained;

wherein MFAP2=gene expression level of MFAP2;

wherein ECOG=ECOG performance status; and

wherein VEGFA=gene expression level of VEGFA.

3. The method of Embodiment 1 or 2, the method further comprising calculating the patient's risk of recurrence at time t (λ(t)) wherein

λ( t )= A 0 ( t ) e recurrence_score

wherein λ 0 (t) is the baseline hazard function estimated with a non-parametric strategy.

4. The method of any one of the preceding Embodiments, wherein determining the expression levels of MFAP2 and VEGFA comprises measuring levels of RNA transcripts

5. The method of Embodiment 4, wherein the method further comprises normalizing the level of the RNA transcripts to obtain normalized gene expression levels.

6. The method of any one of the preceding Embodiments, wherein the biological sample containing cancer cells is fixed, paraffin-embedded, fresh, or frozen.

7. The method of any one of the preceding Embodiments, wherein the method further comprises computing the patient's risk of recurrence at time t if the patient receives platinum-based therapy.

8. The method of any one of the preceding Embodiments, wherein the method further comprises computing the patient's risk of recurrence at time t if the patient receives bevacizumab.

9. The method of Embodiment 8, wherein the method comprises calculating the benefit of the patient receiving bevacizumab and platinum-based therapy versus platinum-based therapy without bevacizumab.

10. The method of any one of the preceding Embodiments, wherein the method further comprises administering bevacizumab or platinum-based therapy or both.

11. The method of Embodiment 10, wherein the method comprises administering bevacizumab only if the patient's risk of recurrence at time t of the patient receiving bevacizumab is greater than the patient's risk of recurrence at time t of the patient receiving platinum-based therapy without bevacizumab.

12. The method of Embodiment 11, wherein the difference in the patient's risk of recurrence at time t is at least 1 month, at least 2 months, at least 3 months, at least 4 months, at least 5 months, or at least 6 months.

Exemplary Method Embodiments for Predicting the Response of a Patient with Ovarian Cancer to Treatment with Bevacizumab

1. A method for predicting the response of a patient with ovarian cancer to treatment with bevacizumab, the method comprising:

determining gene expression levels of VEGFA and MFAP2; calculating a FIGO numeric score, wherein the FIGO stage is coded as an integer; calculating a surgical outcome score, wherein the score is −1 if the surgical outcome was suboptimal; 0 if the surgical outcome was optimal but tumor tissue smaller than 1 cm remained; or +1 if the surgical outcome was optimal and no visible macroscopic tumor tissue remained; calculating an ECOG score of 0 to 2, based on ECOG performance status; applying the expression levels, FIGO numeric score, surgical outcome score, and ECOG score to a predictive model that relates the variables with progression-free survival of ovarian cancer; and evaluating an output of the predictive model to predict progression-free survival of the patient.

2. The method of Embodiment 1, wherein the method further comprises applying the expression levels, FIGO numeric score, surgical outcome score, and ECOG score to a predictive model that relates the variables with progression-free survival of a patient with ovarian cancer if the patient is given platinum-based therapy or with progression-free survival of a patient with ovarian cancer if the patient is given platinum-based therapy and bevacizumab.

3. The method of any one of the preceding Embodiments, wherein the predictive model comprises a Cox model.

4. A method for predicting the response of a patient with ovarian cancer to treatment with bevacizumab, the method comprising:

determining gene expression levels of a collection of genes taken from a biological sample of the patient, wherein the collection of genes comprises at least 80%, at least 90%, at least 95%, at least 98%, or 100% of the genes of any one of Tables 9-12; applying the expression levels to a predictive model that relates the expression levels of the collection of genes the likelihood of progression-free survival of the patient; and evaluating an output of the predictive model to predict the likelihood of progression-free survival of the patient.

5. The method of Embodiment 4, wherein the collection of genes is selected from the genes of any one of Tables 9-12 by optimizing the predictive performance with a constraint.

6. The method of Embodiment 4 or 5, the method further comprising applying at least one of FIGO stage, surgical outcome, ECOG score, and tumor histology to the predictive model.

7. The method of any one of Embodiments 4 to 6, wherein the expression levels of the collection of genes are determined at multiple times.

8. The method of any one of Embodiments 4 to 7, wherein the biological sample is fixed, paraffin-embedded, fresh, or frozen.

9. The method of any one of the preceding Embodiments, wherein the predictive model calculates progression-free survival of a patient with ovarian cancer if the patient is given platinum-based therapy and progression-free survival of a patient with ovarian cancer if the patient is given platinum-based therapy and bevacizumab.

10. The method of any one of the preceding Embodiments, wherein the predictive model comprises a support vector machine model.

11. A method comprising the method of any one of the preceding Embodiments and further comprising administering platinum-based therapy or bevacizumab or both to the patient.

Exemplary Method Embodiments for Predicting the Progression-Free Survival of a Patient with Ovarian Cancer

1. A method for predicting progression-free survival of a patient with ovarian cancer, the method comprising:

determining gene expression levels of a collection of genes taken from a biological sample of the patient, wherein the collection of genes comprises at least 80%, at least 90%, at least 95%, at least 98%, or 100% of the genes of any one of Tables 6, 7, or 13-68; applying the expression levels to a predictive model that relates the expression levels of the collection of genes with progression-free survival of ovarian cancer; and evaluating an output of the predictive model to predict progression-free survival of the patient.

2. The method of Embodiment 1, wherein the collection of genes is selected from the genes of any one of Tables 6, 7, or 13-68 by optimizing the predictive performance with a constraint.

3. The method of Embodiments 1 or 2, the method further comprising applying at least one of FIGO stage, surgical outcome, and tumor histology to progression-free survival of a patient with ovarian cancer.

4. The method of any one of the preceding Embodiments, the method further comprising detecting an additional biomarker of progression-free survival of the patient.

5. The method of Embodiment 4, wherein the additional biomarker of progression-free survival comprises a germline mutation, a somatic mutation, a DNA methylation marker, a protein marker, or a combination thereof.

6. The method of any one of the preceding Embodiments, wherein the expression levels of the collection of genes are determined at multiple times.

7. The method of any one of the preceding Embodiments, wherein the predictive model comprises a support vector machine model.

8. The method of any one of the preceding Embodiments, wherein the biological sample is fixed, paraffin-embedded, fresh, or frozen.

9. A method comprising the method of any one of the preceding Embodiments and further comprising administering platinum-based therapy or bevacizumab or both to the patient.

Exemplary Method Embodiments for Predicting an Outcome for a Patient with Ovarian Cancer

1. A method for predicting an outcome for a patient, the method comprising:

receiving an identified set of biomarkers determined based on a set of predetermined data comprising clinical data, gene expression data, or both; identifying other sets of biomarkers based on the identified set of biomarkers and remaining data comprising the set of predetermined data excluding the identified set of biomarkers; generating a signature for each set of biomarkers to predict an outcome for a patient having ovarian cancer; and determining a prediction of an outcome for a patient having ovarian cancer based on one or more signatures and patient test data comprising clinical data, gene expression data, or both.

2. The method of Embodiment 1, wherein the outcome relates to progression-free survival for a patient with ovarian cancer.

3. The method of Embodiment 1, wherein the outcome relates to benefitting from the administration of bevacizumab, platinum-based chemotherapy, or both for a patient with ovarian cancer.

4. The method of any one of the preceding Embodiments, wherein generating a signature for each set of biomarkers comprises feeding each set of biomarkers into a machine learning classifier fitting and model pipeline.

5. The method of Embodiment 4, wherein the machine learning classifier fitting and model pipeline incorporates model selection and error estimation.

6. The method of Embodiment 4 or 5, wherein the machine learning classifier fitting and model pipeline applies one or more of the following: a repeated nested n-fold cross validation with grid parameter choice, a support vector machine classifier, a random forest classifier, or a lasso classifier.

7. The method of any one of the preceding Embodiments, wherein determining a prediction of an outcome for a patient having ovarian cancer is based on an ensemble prediction using one or more signatures.

8. The method of Embodiment 7, wherein the ensemble prediction averages outputs of each signature.

9. The method of Embodiment 7, wherein the ensemble prediction uses one or more signatures or each signature based on available patient test data.

10. The method of any one of the preceding Embodiments, wherein each signature is statistically indistinguishable from another signature for a particular predictivity level.

11. The method of any one of the preceding Embodiments, wherein each signature is a minimal set of biomarkers for a particular predictivity level.

12. The method of any one of the preceding Embodiments, wherein each signature comprises some or all genes of any of Tables 6, 7, or 9-68.

13. The method of any one of the preceding Embodiments, wherein identifying other sets of biomarkers comprises feeding the identified set of biomarkers and remaining data into a TIE* algorithm to provide the other sets of biomarkers.

14. The method of Embodiment 13, wherein the TIE* algorithm identifies the Markov Boundary set of biomarkers in the remaining data.

15. The method of Embodiment 13 or 14, wherein the TIE* algorithm recursively identifies the Markov Boundary sets of biomarkers for different subsets of remaining data.

16. A method for predicting an outcome for a patient, the method comprising:

determining a prediction of an outcome for a patient having ovarian cancer based on one or more signatures and patient test data comprising clinical data, gene expression data, or both, wherein the one or more signatures are generated to be statistically indistinguishable from a signature of any one of Tables 6, 7, or 9-68 for predicting a clinical response to bevacizumab, platinum-based chemotherapy, or both.

17. A method comprising the method of any one of the preceding Embodiments and further comprising administering platinum-based therapy or bevacizumab or both to the patient.

The present invention is illustrated by the following examples. It is to be understood that the particular examples, materials, amounts, and procedures are to be interpreted broadly in accordance with the scope and spirit of the invention as set forth herein.