SEMINAR: Efficient alternatives to foundation models for single-cell and spatial transcriptomics

Guest: Mehmet Gönen, Koç University

Title: Efficient alternatives to foundation models for single-cell and spatial transcriptomics (BIO, CS, DS)

Date/Time: December 30, 2025, 13:40

Location: FENS L055

Abstract: This talk explores emerging challenges in computational biology driven by the rise of deep learning foundation models. Researchers increasingly rely on these models rather than designing customized solutions.
 

First, I will examine how single-cell foundation models are applied to downstream tasks in computational biology. Current workflows rely heavily on linear dimensionality reduction techniques that often fail to resolve complex non-linear biological signals. While deep learning foundation models address this limitation, they introduce prohibitive training costs and hardware dependencies. To address these problems, we evaluated Random Fourier Features (RFF) as a scalable and mathematically principled alternative. RFF approximates non-linear kernels through a randomized feature mapping, enabling fast linear algorithms to capture complex non-linear relationships without the need for model training. We showed that kernel approximations offer a powerful alternative for high-dimensional single-cell analysis.
 

Next, I will present DOST, a customized machine learning algorithm for spatial transcriptomics (ST). Existing computational methods for identifying spatial domains often rely on computationally expensive Bayesian inference or deep learning models. That is why we developed DOST, a distance-preserving and optimization-based method that clusters ST spots into spatial domains using an explainable loss function combining gene expression similarity with spatial organization. Across diverse ST technologies, DOST delivers competitive or superior clustering accuracy compared to recent methods (BASS, ADEPT, GraphST), while dramatically reducing runtime.

Bio: Mehmet Gönen, received the B.Sc. degree in industrial engineering, the M.Sc. and the Ph.D. degrees in computer engineering from Boğaziçi University in 2003, 2005, and 2010, respectively. He did his postdoctoral work at the Helsinki Institute for Information Technology and the Department of Information and Computer Science, Aalto University. He then worked as a senior research scientist at the Fred Hutchinson Cancer Research Center in Seattle, Washington and as an assistant professor at the Department of Biomedical Engineering, Oregon Health & Science University in Portland. He joined the College of Engineering, Koç University in September 2015 and the School of Medicine, Koç University in December 2016 as an assistant professor. He has been working as a full professor since February 2023. His research interests mainly include machine learning and computational biology. His research has been awarded by the Science Academy of Turkey in 2016, by the Turkish Academy of Sciences in 2017, and by the Health Institutes of Turkey in 2020 under their young scientist award programs.