🔬 Training Transformers to solve 95% failure rate of Cancer Trials — Ron Alfa & Daniel Bear, Noetik
TL;DR
Noetik is tackling the 95% failure rate of cancer clinical trials by training transformers on proprietary multimodal patient tumor data to identify hidden biological subtypes and match therapies to responsive populations, moving beyond simplistic biomarkers and outdated cell lines.
🎯 The Patient Selection Problem 3 insights
Cancer trials fail due to patient mismatch, not drug quality
Ron Alfa argues that 90-95% of oncology drug failures stem from inability to identify responsive patient subpopulations rather than pharmacological flaws in the molecules themselves.
Traditional biomarkers are too simplistic for complex biology
Current clinical methods rely on single mutations or protein stains that miss the rich multimodal patterns determining therapeutic response.
True cancer subtypes remain largely unknown
Pathology-based classifications like 'lung cancer' mask multiple distinct functional subtypes that require data-driven approaches to identify properly.
🔬 Data Generation Strategy 3 insights
Building proprietary datasets from fresh human tumors
Noetik built an in-house lab to process thousands of patient samples into spatial arrays, rejecting decades-old immortalized cell lines that fail to reflect actual human tumor biology.
Intentional dataset design beats brute force collection
Following the Protein Data Bank and ImageNet models, they curate high-quality, multimodal data at scale rather than cobbling together existing public repositories.
Imaging enables scalable, information-dense profiling
Spatial array imaging captures many patients per slide, providing rich visual biological data at significantly lower cost than sequencing runs.
🤖 AI Architecture & Applications 3 insights
Self-supervised transformers learn biological subtypes unbiased
Models identify therapeutically relevant cancer subtypes directly from patient data without preconceptions about whether drivers are genetic, immune, or spatial.
Dual-use models for discovery and trial rescue
The same architecture enables reverse translation for target discovery and retrospective analysis of failed Phase 2/3 trials to design better patient cohorts.
Scaling toward generalizable cancer models
With several hundred patients per major indication, models generalize across cancer types, though Daniel Bear notes biology remains complex and requires continued data expansion.
Bottom Line
Pharmaceutical companies must abandon outdated cell lines and simplistic single-biomarker approaches in favor of training transformers on large-scale, multimodal patient tumor datasets to identify true biological subtypes and match therapies to responsive populations before initiating trials.
More from Latent Space
View all
🔬 "The Most Innovative Diffusion Research Is Happening in Drug Discovery, Not Image Generation"
Evan Fineberg and Sergey Udov of Genesis Molecular AI discuss how diffusion models have pivoted from image generation to drive breakthroughs in 3D protein structure prediction. They detail how their Pearl model applies LLM-style scaling strategies—including synthetic physics-based training data and inference-time 'thinking'—to solve the historically intractable challenge of predicting how small molecules bind to proteins.
Cooking with OpenAI’s Research Chief: AGI, o1, Evals, and Scaling Laws — Mark Chen
OpenAI Chief Research Officer Mark Chen discusses the company's research philosophy while cooking Korean tofu stew, emphasizing that scaling laws remain robust, reinforcement learning excels in objective domains, and successful research organizations balance top-down vision with bottom-up conviction.
The Agent Cloud: Databricks’ Bet on the Future of AI — Matei Zaharia and Reynold Xin
Matei Zaharia and Reynold Xin detail Databricks' open-source 'Agent Cloud' platform (Omnigen), arguing that standardized protocols and persistent infrastructure—not just better models—will determine which enterprises successfully deploy collaborative, secure AI agents at scale.
AI Security After Codex and Claude Code — Zico Kolter & Matt Fredrikson, Gray Swan
Gray Swan co-founders Zico Kolter and Matt Fredrikson explain why AI systems require a fundamentally different security approach than traditional software, highlighting how their automated red teaming system 'Shade' has begun to outperform human experts at finding model vulnerabilities. They emphasize the urgent need to treat AI agents as inherently untrusted entities capable of correlated failures across the software ecosystem.