In this episode, we dive into the revolutionary work at Turbine Simulated Cell Technologies, where the intersection of artificial intelligence and biology is paving the way for a new era in drug development. Join us as we explore the critical challenges of bias in training data sets and the innovative solutions Turbine employs to combat these issues.
Our expert guests, Bence Szalai, MD, PhD, Research Team Lead, and Istvan Taisz, MD, PhD, Project Manager at Turbine, share their insights into how biases in training data can lead AI models astray. Bence highlights a classic example from machine learning history: a model accidentally trained to identify tanks based on lighting conditions, illustrating the pitfalls of relying on flawed data. He explains how foundational models in biology, particularly those predicting protein interactions and drug effects on the cell signaling level, can similarly suffer from biases that compromise their predictive capabilities.
In the episode, Bence and Istvan discuss Turbine’s groundbreaking benchmark study known as EFFECT (Evaluation Framework for Predicting Efficacy of Cancer Treatment). This innovative framework helps identify and eliminate biases, allowing for a more accurate assessment of model performance in real-world scenarios. Their statistical tool, the “bias detector,” specifically addresses known biases, ensuring that the models yield meaningful predictions that are biologically relevant.
In Silico Clinical Positioning
One of the highlights of the discussion is Turbine’s tailored approach to in silico biomarker discovery. Istvan outlines the importance of creating customized avatars that are specifically designed for the biological question at hand. By simulating drug responses across a vast library of in silico avatars representing real biological samples, Turbine can predict drug efficacy with remarkable accuracy.
The episode also delves into how Turbine enriches its training datasets with patient samples. Istvan discusses their collaboration with AstraZeneca, emphasizing the value of recent patient data that reflects current therapeutic challenges. By integrating this information, Turbine significantly enhances its model’s predictive capabilities, achieving a 40% increase in accuracy for previously unseen cell lines.
Bridging Data and Discovery
With cutting-edge methodologies that address the biases inherent in biological data, Turbine is paving towards building computational solutions that truly understand and model biology. A biology-first approach, that’s augmented by machine learning, and a wet lab in Turbine’s training loop ensure that they talk the same language as fellow scientists working in R&D.
This episode is a must-listen for anyone interested in the forefront of biotechnology and drug development. Tune in to learn how Turbine is revolutionizing the landscape of computational biology and improving the efficacy of cancer treatments.
To learn more about Turbine Simulated Cell Technologies and connect with their team, visit www.turbine.ai.