Vinayak Agarwal

Vinayak Agarwal

Principal AI Engineer @ K-Dense | MIT PhD '25 (McDermott Lab, BCS)

Posts by Collection

portfolio

publications

Object-based synthesis of scraping and rolling sounds based on non-linear physical constraints

Published in Proceedings of the 24th International Conference on Digital Audio Effects (DAFx20in21), 2021

Download paper here

Recommended citation: Agarwal, V., Cusimano, M., Traer, J., & McDermott, J. H. (2021). Object-based synthesis of scraping and rolling sounds based on non-linear physical constraints. Proceedings of the 24th International Conference on Digital Audio Effects (DAFx20in21), 136–143.

GeneJepa: A Predictive World Model of the Transcriptome

Published in bioRxiv, 2025

GeneJepa is a predictive world model of the transcriptome developed at K-Dense. It learns representations of gene expression states and predicts downstream transcriptomic responses to perturbations.

Recommended citation: Agarwal, V., Li, O., Kassis, T., Gopinath, A. et al. (2025). GeneJepa: A Predictive World Model of the Transcriptome. bioRxiv. https://biorxiv.org

Revive-Flow: A Foundation Model for Blood DNA Methylation Aging

Published in bioRxiv, 2025

Revive-Flow is a foundation model for blood DNA methylation data, trained to capture aging signatures and biological age acceleration. Developed at K-Dense as part of the longevity AI research program.

Recommended citation: Agarwal, V., Li, O., Kassis, T., Gopinath, A. et al. (2025). Revive-Flow: A Foundation Model for Blood DNA Methylation Aging. bioRxiv. https://biorxiv.org

K-Dense Analyst: Towards Fully Automated Scientific Analysis

Published in arXiv, 2025

K-Dense Analyst is an autonomous AI system that performs end-to-end scientific analysis, compressing months of research workflows into hours or days. It combines multi-agent orchestration, automated hypothesis testing, and rigorous statistical validation.

Recommended citation: Agarwal, V., Kassis, T., Li, O., Gopinath, A. et al. (2025). K-Dense Analyst: Towards Fully Automated Scientific Analysis. arXiv. https://arxiv.org

Transcriptomic Age Prediction Using Mixture-of-Experts Models Reveals Tissue-Specific Aging Signatures

Published in medRxiv, 2025

We develop a mixture-of-experts ensemble clock for transcriptomic age prediction, trained on 57,584 samples spanning 28 tissues (ages 1-114 years). The model achieves R2 = 0.854 and MAE = 4.26 years with calibrated uncertainty intervals, and reveals tissue-specific aging signatures including CDKN2A/p16, AMPD3, MIR29B2CHG, and SEPTIN3.

Recommended citation: Agarwal, V., Li, O., Kassis, T., Gopinath, A. et al. (2025). Transcriptomic Age Prediction Using Mixture-of-Experts Models Reveals Tissue-Specific Aging Signatures. medRxiv. https://medrxiv.org

Limitations of TabPFN for High-Dimensional RNA-seq Analysis

Published in bioRxiv, 2025

We systematically evaluate TabPFN on high-dimensional RNA-seq classification tasks, identifying key limitations in scalability and performance compared to tree-based ensembles and neural baselines, and provide practical guidance for transcriptomics practitioners.

Recommended citation: Agarwal, V., Li, O., Kassis, T. et al. (2025). Limitations of TabPFN for High-Dimensional RNA-seq Analysis. bioRxiv. https://biorxiv.org

talks

teaching

Teaching experience 1

Undergraduate course, University 1, Department, 2014

This is a description of a teaching experience. You can use markdown like any other post.

Teaching experience 2

Workshop, University 1, Department, 2015

This is a description of a teaching experience. You can use markdown like any other post.