AI target discovery engine
Applying AI to multimodal patient data to identify the best candidate targets in a given indication
Challenge
Targets don’t convert to drugs
- Patient data is underutilized
- Pre-clinical models lack translatability to human biology
- Disease heterogeneity is not captured
Solution
Owkin’s TargetMATCH
- TargetMATCH is a suite of end-to-end tools that uses multimodal patient data as the input, and outputs the top candidate targets and paired patient subgroups that would most benefit from therapeutic intervention on these targets
Learn how TargetMATCH identifies the best drug targets and pairs them with patient subgroups [2:37]
Using AI for target discovery
How does TargetMATCH work?
Best-in-class data
Multimodal patient data: Private datasets
Large cancer patient cohorts from partner hospitals with:
- Deep clinical data
- Latest SoC response
- 8 million data points per patient
- 6 data modalities
Multimodal patient data: Public datasets
Tens of thousands of patients from:
- Pan-cancer public datasets (TCGA)
- Disease-specific datasets (MESOMICS)
Pharma data from partners
We can enrich the discovery dataset by integrating proprietary datasets.
Biomedical literature
We use publicly available datasets and biomedical resources to characterize drug targets.
- OpenTargets, GTEX, CT.gov, ChEMBL, and more.
AI characterization
Integration of multimodal data
We use AI to Integrate multiple data modalities in a unified environment.
Patient characterization
We use representation learning to extract features from millions of data points from different data modalities:
- H&E, WES, Clinical data, RNA-seq, single-cell RNA-seq, Spatial transcriptomics
- 4,000 AI features per patient
Target characterization
Targets are represented by vectors of features extracted from public biomedical resources:
- 50 AI features per target
AI reasoning
Target/patient pairing
Our AI builds target-patient pairs by computing the importance of a target in a subpopulation along many axes e.g:
- Patient survival
- Essentiality
- Features of the tumor microenvironment (TME)
Target/patient subgroup pair ranking
We find the best patient subgroup for every target using an innovative AI optimization method that answers the question:
- Which patient subgroup would best respond to targeting gene X?
Biomedical review
- In-house team of MDs and PhDs in biology and pharmacology
- Top external KOLs and clinicians in the field
Discovery
- Top 5 targets and associated patient subgroups
- Actionable biomarkers for clinical success
AI target discovery
Why work with Owkin?
Access to patient data and samples
Access to continuously enriched, up-to-date patient data representing the latest standard-of-care.
Multimodal AI
Experts in multimodal AI analysis to capture disease heterogeneity and the tumor microenvironment for a multiscale understanding of inter-patient response variability.
Interpretable AI
Allows medical experts to critically examine the model rationale and generate new hypotheses to continuously improve models.
Proven track record
Owkin has delivered 3 new AI-discovered candidate drug targets to Sanofi
"I'm excited about the Sanofi and Owkin collaboration and its potential to transform drug discovery. Owkin's data network and AI capabilities combined with Sanofi's expertise, can potentially lead to new treatments and better patient outcomes."
Frank Nestle
Global Head of Research and Chief Scientific Officer, Sanofi
Subgroup discovery
Multimodal AI-powered biomarkers
We combine cutting-edge machine learning and biology to identify biomarkers.
AI drug positioning
Matching the right drug and patient for better responses
For a given drug, we identify novel disease indications and subgroups for development, by aggregating causal evidence from prior knowledge and multimodal patient data.