Seminar • Bioinformatics • Trusted AI for Molecular Inference: The Role of Sparsity

Monday, March 14, 2022 11:30 am - 11:30 am EDT (GMT -04:00)

Please note: This seminar will be given online.

Amirali Aghazadeh, Postdoctoral researcher
Department of Electrical Engineering and Computer Sciences
University of California, Berkeley

Recent breakthroughs in artificial intelligence (AI) have enabled accurate prediction of protein structures from their sequences and have opened up new avenues for the engineering of proteins, drugs, and molecules with advanced and novel functional properties. However, despite their high predictive power, AI models do not provide a mechanistic understanding of interactions that give rise to many functional properties. Moreover, their generalization power has remained limited for novel and rapidly evolving molecules for which sufficient sequence data is not available.

In this talk, I will describe how we develop a foundation for trusted AI in molecular inference. Key to my approach is the observation that the combinatorial landscapes of molecular properties reside in low dimensional subspaces characterized by sparse high order non-linear interactions. I will show how we can leverage this sparsity prior and develop new algorithms rooted in signal processing, graph theory, and large-scale optimization to efficiently explain, regularize, and build molecular AI models. My algorithms have resulted in a drastic reduction in the number of sequences required to infer functional properties in proteins and an improved understanding of high order interactions in the DNA repair process. I will conclude by describing how my works set the computational and statistical foundation for engineering programmable molecular machines.


Bio: Amirali Aghazadeh is a postdoctoral researcher in the Electrical Engineering and Computer Science department at the University of California, Berkeley, working with Kannan Ramchandran. Prior to that, he was a postdoctoral researcher at Stanford University after receiving his PhD degree in Electrical and Computer Engineering from Rice University with Richard Baraniuk. His research interest is at the interface of large-scale machine learning, signal processing, and molecular engineering. He is the recipient of the Hershel M. Rich Invention Award for his thesis on universal molecular diagnostics as well as the Texas Instruments Fellowship. He received his Bachelor’s degree in Electrical Engineering from Sharif University of Technology.


To join this seminar on Zoom, please go to https://uwaterloo.zoom.us/j/92324267678.

Please note: The passcode will be provided by email a week before the seminar as well the morning of the seminar.