Geometry and Physics for Machine Learning in Biomedicine
The MEGA group has a two-pronged mission:
Celcomen leverages a mathematical causality framework to disentangle intra- and inter- cellular gene regulation programs in spatial transcriptomics and single-cell data through a generative graph neural network. It can learn gene-gene interactions, as well as generate post-perturbation counterfactual spatial transcriptomics, thereby offering access to experimentally inaccessible samples. GitHub
We extend the 2d topological gravity model of Marolf and Maxfield to have as its bulk action any open/closed TQFT obeying Atiyah’s axioms. The holographic dualsof these topological gravity models are ensembles of 1d topological theories with random dimension. Specifically, we find that the TQFT Hilbert space splits into sectors, between which correlators of boundary observables factorize, and that the corresponding sectors of the boundary theory have dimensions independently chosen from different Poisson distributions. As a special case, we study in detail the gravity model built from the bulk action of 2d Dijkgraaf-Witten theory, with or without end-of-the-world branes, and for arbitrary finite group G. The dual of this Dijkgraaf-Witten gravity model can be interpreted as a 1d topological theory whose Hilbert space is a random representation of G and whose aforementioned sectors are labeled by the irreducible representations of G.
Models of Virtual Cells and Virtual Tissues at single-cell resolution would allow us to test perturbations in silico and accelerate progress in tissue and cell engineering. However, most such models are not rooted in causal inference and as a result, could mistake correlation for causation.We introduce Celcomen, a novel generative graph neural network grounded in mathematical causality to disentangle intra- and inter-cellular gene regulation in spatial transcriptomics and single-cell data. Celcomen can also be prompted by perturbations to generate spatial counterfactuals, thus offering insights into experimentally inaccessible states, with potential applications in human health. GitHub
We develop EmptyDropsMultiome, an approach that distinguishes true nuclei-containing droplets from background in single-cell multiome data. On real datasets, we observe that CellRanger-arc misses more than half of the nuclei identified by EmptyDropsMultiome and, moreover, is biased against certain cell types, some of which have a retrieval rate lower than 20%.. GitHub
We develop an efficient method for computing thermal partition functions of weakly coupled scalar fields in AdS. We consider quartic contact interactions and show how to evaluate the relevant two-loop vacuum diagrams without performing any explicit AdS integration, the key step being the use of Källén-Lehmann type identities. This leads to a simple method for extracting double-trace anomalous dimensions in any spacetime dimension, recovering known first-order results in a streamlined fashion.
PhD Student
University of Tehran (BSc)
BSc Intern
Sapienza University of Rome (BSc)
External Collaborator, PhD Student at UCLA
University of Cambridge (MSc)
University of Washington (BSc)
We are currently seeking enthusiastic researchers at any levels especially: Postdoc, PhD, MSc, BSc
If you're interested in joining our team, please reach out!