Bayesian AI to delineate molecular signatures of patient susceptibility to potential hematologic events in a phase I study of BPM31510 (ubidecarenone) in solid tumors.

e14042 Background: BPM 31510 is a ubidecarenone containing nanodispersion elicits anticancer effect by switching cancer cells energy generation from glycolysis to mitochondrial OXPHOS; i.e. reverses Warburg effect. This Phase 1 study is evaluating safety of 144-hour infusion of BPM 31510 as a single agent and in combination with 3 standard chemotherapy regimens. The molecular adaptive study design enabled longitudinal multi-omic molecular profiling to delineate personalized cause-and-effect networks relating patient biology to clinical outcomes using unbiased Bayesian statistics based bAIcis Artificial Intelligence (AI) technology Methods: Eligible pts (aged ≥ 18 y) were relapsed/refractory to standard therapy. The monotherapy arm received IV BPM31510 for 144-hour infusion in 28d cycles, and combination arms (gemcitabine, 5-FU or docetaxel) were primed with BPM 31510 for 3 wks followed by weekly dosing in a 6wk cycle. Doses were escalated in a 3+3 schema. Endpoints were safety, PK and multi-omics analysis. Clinical information was processed and integrated with multi-omic molecular data (metabolomics, lipidomics and proteomics), utilized bAIcis to learn relationships and extract actionable information from pt records, molecular features and corresponding clinical response. Results: As of 10 Dec 2016, 97 pts have been enrolled. Hematologically-related DLTs were reported at 171mg/kg in monotherapy and 137mg/kg in the BPM31510+gemcitabine arm (maximum administered dose). The bAIcis-generated set of predictors defining coagulation-related events in the clinical trial population include proteins, lipids, and metabolites, the levels of which can potentially predict undesirable events prior to and 24h after the 1st treatment with BPM 31510. Conclusions: BPM31510 is well tolerated as a monotherapy and in combination with chemotherapeutic agents. Molecular signatures of susceptibility have been identified through use of multi-omic based AI analysis. Molecular signatures predictive of coagulation-related events will be assessed in Phase 2 and 3 trials to guide the development plan for BPM 31510 as anticancer agent. Clinical trial information: NCT01957735.

Adaptive Study Design Through Semantic Association Rule Analysis

Association mining aims to find valid correlations among data attributes, and has been widely applied to many areas of data analysis. This paper presents a semantic network-based association analysis model including three spreading activation methods. It applies this model to assess the quality of a dataset, and generate semantically valid new hypotheses for adaptive study design especially useful in medical studies. The approach is evaluated on a real public health dataset, the Heartfelt study, and the experiment shows promising results.