Assessment of Cellworks Omics Biosimulation Therapy Response Predictions for Patients with Acute Myeloid Leukemia (AML) Using Cellworks Singula™: Mycare-020-01

Background. In addition to clinical considerations (e.g., age, de novo vs secondary disease, comorbidities), therapy selection for AML patients is often based on information considering only cytogenetics and/or molecular aberrations and ignoring other patient-specific omics information that could potentially enable selection of more effective treatments. In turn, despite using cytogenetic and molecular-risk stratification, the current overall outcome of AML patients remains relatively poor. The Cellworks Singula™ report predicts clinical response to physician-prescribed treatments using the novel Cellworks Omics Biology Model (CBM) that simulate in silico downstream molecular effects on cell signaling and survival of drug treatments in patient-specific diseased cells.

Methods. The performance of Singula™ was evaluated in a cohort of 474 AML patients aged 2 to 85. The pre-defined Singula™ Classifier utilizes individual patients’ next-generation sequencing (NGS) profiles to provide a dichotomous prediction of response or non-response to the physician prescribed treatments. The clinical outcome data for these subjects, i.e., complete response (CR) and overall survival (OS), were obtained from the TCGA and other 144 PubMed publications, each including also information on patients’ cytogenetics, targeted gene mutations, and/or whole exome sequencing. Blinded to clinical outcomes, Cellworks utilized the cytogenetic and molecular data to generate a Singula™ predicted response (i.e., CR vs non-response) classification for each patient. Statistical analyses, including assessments of accuracy, sensitivity, specificity, and negative (NPV) and positive predictive (PPV) values were performed to compare the Singula™ predicted clinical response to the actual observed clinical response. Kaplan-Meier curves, associated log rank tests and median OS are provided for patients stratified by Singula™ predicted response. Multivariate Cox proportional hazards regression was used to further test the hypothesis that Singula™ is an independent predictor for OS once adjusted for patient age and actual prescribed treatment.

Results. Data are summarized in Table 1. The Singula™ classifier had 92.3% (90.6%, 95.3%) accuracy in predicting correctly observed patient complete response to the prescribed treatment. with 97.3% (95.0%, 98.8%) sensitivity. Singula™ had 83.3% (76.1%, 89.1%) specificity for the non-responder patients (n=138; 29.1%). For each of the non-responders, Singula™ provided an alternative treatment therapy predicted to produce clinical response. Assuming at least 2% of the non-responders would have responded to the alternative Singula™ prescribed treatment, Singula™ improves CR rates compared to the original physician prescribed treatment (McNemar’s p‑value < 0.05). Figure 1 provides the Kaplan-Meier curves of Singula-predicted responders vs non-responders for a subset of 292 subjects that had OS data available. In multivariate Cox proportional hazards models, the Singula Classifier remained a significant predictor of overall survival (HR = 2.171, p-value < 0.0001) once adjusted for patient age and physician prescribed treatment.

Conclusions. Cellworks Singula™ has high accuracy and sensitivity in predicting CR for AML patient. Singula also has high specificity in identifying patients who are unlikely to respond physician and may prescribed potentially effective therapies. The Singula™ predicted responders have a significantly longer OS than the predicted non responders. Thus, Cellworks Singula™ can accurately predict AML response, be used to validate or reject a physician’s therapy selection decision and, eventually, provide alternative treatment recommendations.