Karush Suri

Title: Karush Suri and Machine Learning for Molecular-Phenomic Discovery.

Introduction

Karush Suri is an inventor based in Montreal, Canada, whose patent work explores how machine learning can connect molecular structures with cellular phenomic images. His disclosed invention focuses on generating joint-space embeddings that can help researchers interpret the effects of molecules on cellular functions.

Latest Patents

Suri is associated with 1 patent, titled “Utilizing Contrastive Machine Learning Models to Extract Joint-Space Molecular-Phenomic Embeddings from Molecular Structures or Phenomic Images.” The invention describes systems and methods that use a contrastive model to learn a shared latent space for molecular structures and phenomic images. It combines image embeddings from a pretrained phenomic encoder with molecular structural embeddings, using techniques that include an inter-sample similarity-aware loss, concentration encoding, and undersampling of inactive molecular data.

The resulting embeddings can support several types of molecular inference, including identifying similar molecules or phenomic images, estimating the phenotypic impact of particular molecules, and classifying molecular activity.

Career Highlights

Suri works at Recursion Pharmaceuticals, Inc., a company whose research applies computational methods to biological discovery. His patent reflects an approach to organizing molecular and image-based data in a common representation, with the aim of enabling more systematic analysis of molecular effects.

Collaborations

Suri’s listed coworker is Jan Frederik Wenkel. The patent’s described methods bring together molecular structures, phenomic images, and machine-learning techniques to support biological and molecular analysis.

Conclusion

Karush Suri’s patented work applies contrastive machine learning to connect molecular structures with phenomic observations, creating embeddings that can support a range of molecular inferences.