Integrating AI and Machine Learning in Life Sciences - Allen School Colloquia

Explore cutting-edge research in AI and machine learning applications for life sciences in this 53-minute Allen School Colloquium featuring the AIMS Research Group. Dive into recent advancements in explainable AI, computational biology, and medicine, including COVID-19 detection in medical imaging, synergistic drug combinations for cancer treatment, and contrastive latent variable modeling for biological discovery. Learn about the challenges and opportunities in integrating AI/ML with life sciences, from developing interpretable models to identifying causes and treatments for diseases like cancer and Alzheimer's. Gain insights from PhD students and researchers as they discuss their work on efficient algorithms, theoretical foundations, and practical applications of AI in high-stakes domains such as healthcare and precision medicine.

Intro
Al for Bio-Medical Sciences (AIMS) Lab
Overview
Published Al models detect COVID-19 in chest X-rays
What was the training data for these published models?
How can we test how robust the models are?
How robust are the models?
What is important for the model's predictions?
Can we fix shortcut learning with improved data?
Conclusions
Acknowledgements
Outline
Explainable AI
Examples
Intuition
The recipe
Unified framework
SHAP's ingredients
Human-friendly
Game-theoretic
Information-theoretic
Choosing optimal combinations is hard
This is the perfect opportunity for predictive models
In high stakes scenarios, models should be interpretable
A simple fix: ensemble attributions
Interpretability uncovers transcriptional programs
Bringing Interpretable Models to Cancer Precision Medicine
Contrastive Analysis
Latent Variable Models
VAE Model
Problem
Contrastive Latent Variable Model
Background datasets
Contrastive VAE
Inspecting the salient latent values