🔬 Training Transformers to solve 95% failure rate of Cancer Trials — Ron Alfa & Daniel Bear, Noetik

2026-04-15 · Source: Latent.Space - Www.latent.space · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Life Sciences & Biology, Pharmaceuticals & Biotechnology · Depth: Advanced, extended

Summary

Noetik, a company co-founded by Ron Alfa and Dan Baer, is addressing the 90-95% failure rate of cancer drugs in clinical trials by focusing on patient selection rather than drug pharmacology or target selection. Their thesis posits that drugs fail because we are poor at identifying which patients will respond. Noetik generates extensive multimodal patient data, including H&E pathology, protein stains, and spatially resolved RNA (spatial transcriptomics), to train autoregressive transformer models like TARIO-2. These models aim to understand patient biology and identify therapeutically relevant cancer subtypes, enabling the selection of appropriate patient populations for clinical trials and potentially rescuing failed trials. The company emphasizes generating high-quality, curated datasets at scale, similar to ImageNet, to overcome limitations of existing public repositories and traditional drug development methods that rely on non-representative cell lines and animal models.

Key takeaway

For Directors of AI/ML in biopharma seeking to improve clinical trial success rates, Noetik's approach highlights that investing in large-scale, high-quality multimodal patient data generation and advanced transformer models can significantly enhance patient stratification. Your teams should explore integrating similar data-centric AI strategies to move beyond traditional, less predictive biomarkers and identify precise patient cohorts, thereby increasing the probability of drug efficacy and reducing trial failures.

Key insights

Improving cancer drug success requires advanced AI models for precise patient selection based on multimodal biological data.

Principles

• Drug failure often stems from poor patient selection.
• Multimodal patient data is crucial for understanding disease subtypes.
• Data scale and quality are paramount for effective AI in biology.

Method

Noetik trains autoregressive transformer models on proprietary, large-scale multimodal patient data (H&E, protein stains, spatial transcriptomics) to identify therapeutically relevant cancer subtypes and simulate drug responses for patient stratification.

In practice

• Use H&E images for diagnostic prediction of drug response.
• Simulate gene knockdowns to predict drug efficacy.
• Leverage spatial transcriptomics for detailed molecular insights.

Topics

• Cancer Clinical Trials
• Patient Selection
• Multimodal Biomedical Data
• Autoregressive Transformers
• Spatial Transcriptomics

Best for: AI Scientist, Research Scientist, Director of AI/ML

Related on AIssential

• AI Improving Dose Decisions and Patient Outcomes in Oncology- with Shefali Kakar of Novartis — AI-driven data integration refines oncology dose decisions, safety, and patient variability, accelerating drug development.
• Simulating clinical interventions with a generative multimodal model of human physiology — HealthFormer is a generative transformer modeling human physiology for forecasting, risk stratification, and intervention simulation.
• Multi-modal AI for comprehensive breast cancer prognostication — A multi-modal AI test significantly improves breast cancer prognostication by integrating pathology and clinical data.
• Skill-Augmented AI Agents for Medical Research Analysis: An Exploratory Multi-Model Human Evaluation in an NSCLC Transcriptomic Biomarker Task — Skill-augmented AI agents showed a directional, non-confirmatory quality improvement in medical research analysis tasks.
• GigaTIME: Scaling tumor microenvironment modeling using virtual population generated by multimodal AI — GigaTIME uses multimodal AI to convert common H&E slides into virtual mIF images, enabling population-scale precision immuno-oncology research.
• AI identifies early risk patterns for skin cancer — AI models can predict melanoma risk with high accuracy using routine healthcare data, enabling targeted screening.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Latent.Space - Www.latent.space.