[CLSTR]
< Back to all clusters
[TECHNOLOGY] · United Kingdom, Israel, Switzerland · 5 sources

started · updated

Shift Bioscience introduces new framework for AI virtual cell benchmarking

AI virtual cells drug discovery clinical trial simulation genetic perturbation models biotechnology

Shift Bioscience has published a study in Nature Biotechnology introducing a new framework to evaluate AI virtual cells used in drug target discovery. The research addresses concerns that deep learning models for predicting genetic perturbations often failed to outperform simple baselines in previous studies.

The company argues that earlier benchmarks suffered from miscalibrated metrics that lacked the sensitivity to detect meaningful biological signals. By implementing a new benchmarking system, including a positive-control baseline and the dynamic range fraction (DRF), the study demonstrates that well-calibrated virtual cell models can generate biologically significant insights and outperform existing baselines.

Separately, the broader pharmaceutical industry is increasingly adopting AI simulations to mitigate the high costs of clinical trial failures. Companies like BioinvestGPT and QuantHealth are using virtual trials to predict drug efficacy and safety. For instance, BioinvestGPT’s simulations previously predicted an insignificant clinical benefit for Novartis’ del-desiran, a drug that later missed its trial goals, resulting in a significant loss of market value for the Swiss firm.

Entities

BioinvestGPT · Nature Biotechnology · Novartis · QuantHealth · Shift Bioscience

Claims

What the coverage asserts, and how many sources carry each claim.

  • [○ 1 SOURCE] BioinvestGPT ran a simulated trial in July that predicted an insignificant clinical benefit for Novartis’ experimental drug del-desiran. www.emirates247.com
  • [○ 1 SOURCE] AI companies are using virtual clinical trials to predict drug success and reduce the costs of failed human studies. www.emirates247.com
  • [● 2 SOURCES] The new framework uses well-calibrated metrics to show that virtual cell models can outperform simple baselines. distilinfo.com · www.genengnews.com
  • [● 2 SOURCES] Shift Bioscience published a peer-reviewed study in Nature Biotechnology regarding a framework for evaluating AI virtual cells. distilinfo.com · www.genengnews.com
  • [● 2 SOURCES] Previous benchmarks may have yielded poor results due to miscalibrated comparison metrics that reduced sensitivity to performance differences. distilinfo.com · www.genengnews.com

Sources

Shift Bioscience Publishes AI Virtual Cells Benchmark Study [distilinfo.com]
Anthropic’s announcement of a ‘Claude-led’ CRISPR-like discovery is sparking controversy. Here’s why [krdo.com]
AI companies run virtual drug trials to predict failures, cut costs and boost human study success [www.emirates247.com]
AI companies run virtual human drug trials, aim to improve success of human studies [freerepublic.com]
New Calibration Framework Improves Confidence in AI Virtual Cell Models [www.genengnews.com]