Machine-learning-enabled predictive biomarker discovery and patient stratification using standard-of-care data

Inventors

Probert, Christopher • MCCAW, Zachary Ryan • Koller, Daphne • Shcherbina, Anna

Assignees

Insitro Inc

Abstract

The present disclosure relates generally to biomarker discovery and patient stratification, and more specifically to machine learning techniques for discovering relevant biomarkers using data collected as part of the standard-of-care (SoC), which can be used to identify a relevant patient population for a therapeutic with a known mechanism of action (MoA). An exemplary method for predicting activity of a molecular analyte of a patient comprises: training a first module of a machine learning model based on a plurality of medical images of a first cohort; training a second module of the machine learning model based on one or more molecular analyte data sets obtained from a second cohort; receiving a medical image from the patient; and predicting, using the trained first and second modules of the machine learning model, the activity of the molecular analyte from the medical image of the patient.

Core Innovation

The disclosure addresses predicting activity of a molecular analyte of a patient from a medical image using a machine-learning model trained across two cohorts. A first module is trained based on a plurality of medical images of a first cohort and includes an embedding module, while a second module is trained based on one or more molecular analyte data sets obtained from a second cohort and includes one or more heads.

At inference, the approach receives a medical image from the patient and inputs the medical image into the first module to obtain an embedding. The activity of the molecular analyte is then predicted using the embedding together with the trained second module of the machine-learning model.

The disclosure further specifies that molecular analyte activity can be represented using outputs associated with molecular analyte data sets, and that the model can include additional mechanisms and downstream outputs. Such mechanisms include optional attention and tile-level predictions or annotation maps, and training approaches can include transfer learning and domain adaptation, with downstream modeling linking predicted molecular analyte activity to clinical outcomes to select relevant biomarkers and assign patients to responder or non-responder subgroups.

Claims Coverage

The identified independent claims are clm-00001, clm-00028, and clm-00029. Each covers the two-module machine-learning framework with an embedding module trained on medical images and a second module with one or more heads trained on molecular analyte data sets, followed by inference to predict molecular analyte activity from a patient medical image.

Across clm-00001, clm-00028, and clm-00029, the claim coverage centers on training a first embedding module on medical images from a first cohort and a second module with one or more heads on molecular analyte data sets from a second cohort, then using the resulting embedding plus the trained second module to predict molecular analyte activity for a patient medical image.

Stated Advantages

Documented Applications