Mostofa Rafid Uddin

I am a doctoral student in the CPCB Ph.D. Program in Carnegie Mellon University School of Computer Science, working with Dr. Min Xu. I am also a Center for Machine Learning and Health (CMLH) Fellow in Digital Health Innovation. My research interests include un/self-supervised learning, Representation Learning, Computational Biology, 3D Computer Vision, Explainable AI, Vision Foundation Models.

My PhD thesis involves developing unsupervised algorithms for modelling subcellular structure morphology from 2D and 3D microscopic images. Before joining my PhD, I graduated with B.Sc.Engg. in Computer Science and Engineering from Bangladesh University of Engineering and Technology(BUET) and later worked as a lecturer. During that time, I worked with Dr. Md. Shamsuzzoha Bayzid on leveraging machine translation for protein structure prediction.

Research Details

Unsupervised Multi-scale Segmentation of Cellular cryo-electron Tomograms with Stable Diffusion Foundation Model

Skills: Vision Foundation Models, XAI, Transfer Learning, Image Segmentation, 3D Computer Vision, Cryo-ET.

We introduce an unsupervised approach for segmenting multiscale subcellular objects in 3D volumetric cryo-electron tomography (cryo-ET) images, addressing key challenges such as large data volumes, low signal-to-noise ratios, and the heterogeneity of subcellular shapes and sizes. The method requires users to select a small number of slabs from a few representative tomograms in the dataset. It leverages features extracted from all layers of a Stable Diffusion foundation model, followed by a novel heuristic- based feature aggregation strategy. Segmentation masks are generated using adaptive thresholding, refined with CellPose to split composite regions, and then utilized as pseudo-ground truth for training deep learning models. This fully automated, data-driven framework enables the mining of multi-scale subcellular patterns, paving the way for accelerated biological discoveries from large-scale cellular cryo-ET datasets.

A Generic Unsupervised Approach for Disentangling Semantic Content from Parameterized Transformations.

Skills: Unsupervised Learning, Representation Learning, Inverse Problems.

In many real-life image analysis applications, particularly in biomedical research domains, the objects of interest undergo multiple transformations that alter their visual properties while keeping the semantic content unchanged. Disentangling images into semantic content factors and transformations can provide significant benefits for many domain-specific image analysis tasks. To this end, we propose a generic unsupervised framework, Harmony, that simultaneously and explicitly disentangles semantic content from multiple parameterized transformations. Harmony leverages a simple cross-contrastive learning framework with multiple explicitly parameterized latent representations to disentangle content from transformations. To demonstrate the efficacy of Harmony, we apply it to disentangle image semantic content from several parameterized transformations (rotation, translation, scaling, and contrast). Harmony achieves significantly improved disentanglement over the baseline models on several image datasets of diverse domains. With such disentanglement, Harmony is demonstrated to incentivize bioimage analysis research by modeling structural heterogeneity of macromolecules from cryo-ET images and learning transformation-invariant representations of protein particles from single-particle cryo-EM images. Therefore, Harmony is generalizable to many other imaging domains and can potentially be extended to domains beyond imaging as well.

Project is available here.

Unsupervised Identification of Protein Compositions and Conformations via Implicit Content-Transformation Disentanglement

Skills: Contrastive Learning, Representation Learning, Shape Analysis, Protein Biology.

Identifying different protein compositions and conformations from microscopic images of protein mixtures is a challenging open problem. We address this through disentangled representation learning, where separating protein compositions and conformations in an intermediate latent space enables accurate identification. By modeling compositions and conformations as content and transformation, the task can be reduced to disentangling content and transformation. The existing disentangling methods require an explicit parametric form for the transformation, which is unavailable for conformation, making these methods unsuitable. To overcome this limitation, we propose DualContrast, a novel contrastive learning-based method that implicitly parameterizes and disentangles both transformation and content. DualContrast achieves this by generating positive and negative pairs for content and transformation in both data and latent spaces. We demonstrate that existing self-supervised approaches fail under similar implicit parameterization, underscoring the necessity of our method. Through extensive experiments on 3D microscopic images of protein mixtures and additional shape-focused datasets beyond microscopy, we validate our claims and demonstrate the first in-principle fully unsupervised identification of different protein compositions and conformations in 3D microscopic images.

Project is available here.

Teaching Experiences

Awards & Honors

• CPCB Outstanding Research Accomplishment Award
• CMLH Fellowshiop in Digital Health 2023 news
• Deans List Award and University Merit Scholarship
• 1^st Place - Poster Presentation, International Conference on Networking, Systems and Security (4th NSysS 2017) pdf
• 1^st Place - Hackathon for environmental migrants in Bangladesh arranged by Wageningen University, Netherlands. report
• 2^nd Place - Bracathon 2017 by BRAC
• 3^rd Place - National Hackathon 2016 by ICT Division.
• 3^rd Place - BUET Website Design Competition by IICT, BUET