Akshay Subramanian

I am a 4th year PhD student in Materials Science and Engineering at MIT working on generative models and atomistic simulations for molecules and materials.

I'm currently working under Prof. Rafael Gomez-Bombarelli on computationally designing and studying molecules for organic electronics. I interned last Summer with the Display team at Samsung Semiconductor where I worked on flow matching for accelerated sampling of molecular clusters. Before my PhD, I worked on somewhat different topics at the intersection of scientific literature mining and computer vision with Prof. Gerbrand Ceder's group at Lawrence Berkeley National Lab as a part of my undergrad thesis.

Outside of work, I like running, playing tennis, and music! Feel free to get in touch to chat about research or anything else.

Email  /  Google Scholar  /  LinkedIn  /  GitHub  /  Twitter

profile photo
Research
Symmetry-Constrained Generation of Diverse Low-Bandgap Molecules with Monte Carlo Tree Search
Akshay Subramanian, James Damewood, Juno Nam, Kevin P. Greenman, Avni P. Singhal, Rafael Gomez-Bombarelli
arXiv, 2024
arXiv / code

Constraining molecular generators to follow "symmetry" of patent-mined molecules that are likely to be synthesizable.

Closing the Execution Gap in Generative AI for Chemicals and Materials: Freeways or Safeguards
Akshay Subramanian, Wenhao Gao, Regina Barzilay, Jeffrey C. Grossman, Tommi Jaakkola, Stefanie Jegelka, Mingda Li, Ju Li, Wojciech Matusik, Elsa Olivetti, Connor Coley, Rafael Gomez-Bombarelli
An MIT Exploration of Generative AI, 2024  
project page

Our take on what the "execution gap" is in generative AI for chemistry and materials, and the associated dual-use risks and mitigation strategies.

COVIDScholar: An automated COVID-19 research aggregation and analysis platform
John Dagdelen, Amalie Trewartha, Haoyan Huo, Yuxing Fei, Tanjin He, Kevin Cruse, Zheren Wang, Akshay Subramanian, Benjamin Justus, Gerbrand Ceder, Kristin A. Persson
PLOS ONE, 2023  
search engine / arXiv / code

COVID-19 literature search powered by NLP algorithms.

Automated patent extraction powers generative modeling in focused chemical spaces
Akshay Subramanian*, Kevin P. Greenman*, Alexis Gervaix, Tzuhsiung Yang, Rafael Gomez-Bombarelli
Digital Discovery, 2023  
arXiv / code

Patent extracted molecular datasets enable domain-focused training of generative models.

Dataset of gold nanoparticle sizes and morphologies extracted from literature-mined microscopy images
Akshay Subramanian, Kevin Cruse, Amalie Trewartha, Xingzhi Wang, A. Paul Alivisatos, Gerbrand Ceder
arXiv, 2022  
arXiv / code

Automatic extraction and analysis of microscopy images related to gold nanoparticles synthesis from materials science literature, with computer vision techniques.