Material property & novel material prediction

"What I cannot create, I do not understand." - RP Feynman

Imagine being handed a chemical structure that has never been synthesized before. What properties would it have? What color would it display? Which applications or industries may benefit from it?

These questions drive our research in designing novel materials with specific engineered properties. Material prediction presents a unique inverse challenge - distinct from material property prediction. While a given material structure may have definite properties (like a specific boiling point at a given pressure), the reverse problem (discovering materials exhibiting a specific property) is more complex; Multiple materials could satisfy a desired property. This vast combinatorial space makes our mission particularly challenging and exciting.

Interpretable AI for Science

"Robust, Reduced, and Reusable"

Neural networks: Embrace the power - tame the output

Scientists work in the physical world, which has been modeled remarkably well by analytical expressions. Simultaneously, the uses of neural networks to model scientific data have been expanding, despite historically being reserved for use in areas like computer vision.

Analytical expressions often are too complex to extract from these data domains existing in very high dimensions. However, utilizing neural networks together with symbolic regression allows the recovery of explainable expressions from modeling on the intrinsic dimensionality of physical data domains that may exist in higher dimensional spaces.

Our lab takes advantage of the synergy between symbolic regression on neural network latent spaces - along with more recent advances in language modeling - to provide priors for discovering new equations that govern physical data.

Systems biology and bioinformatics

"Reveal, Refine, Rejuvenate"

Our research leverages advanced computational methods to understand the complex yet consistent physical processes within cellular senescence:

• Analyzing RNA velocity using GflowNets to track cellular state transitions
• Developing predictive models for cellular trajectories
• Creating novel computational tools to interpret large-scale biological data

By combining machine learning with biological insights, we aim to decode biological processes from molecular interactions to tissue-level changes. Our ultimate goal is to combine this work with nanoparticle predictions for theranostics prediction and development.