Master’s Project: Using Large Language Models for Analyzing Single-Cell RNA-seq data

Single-cell RNA sequencing (scRNA-seq) has emerged as a powerful technology for studying gene expression at the individual cell level. The complexity and volume of data generated by scRNA-seq experiments present significant challenges for data analysis and interpretation. Traditional methods struggle to effectively capture the rich information contained within scRNA-seq data, making it necessary to explore innovative approaches. Large Language Models (LLMs), like GPT and BERT, have demonstrated remarkable capabilities in understanding and generating human language. Leveraging these models for scRNA-seq data analysis presents an exciting opportunity to unlock new insights in the field of single-cell genomics.

Responsibilities: The selected master’s student will work closely with a PhD student and will be responsible for:

• Assembling Datasets: Create a dataset of scRNA-seq data, both expression and textual documentations suitable for input to LLMs. This representation should capture the multi-dimensional nature of gene expression profiles at the single-cell level.
• Training and Fine-tuning: Train and fine-tune a deep learning model (e.g., BERT) on a comprehensive scRNA-seq dataset. This process will adapt the model to understand the unique language and structure of single-cell genomics data.
• Feature Extraction: Use the trained LLM to extract relevant features from scRNA-seq data, enabling better characterization of cell types, identification of differentially expressed genes, and discovery of novel biological insights.
• Validation and Benchmarking: Rigorously validate the performance of the LLM-based approach against benchmark datasets and compare it with traditional methods. Assess its ability to identify known biological phenomena and discover novel insights.

Prerequisites:

• Experience in bioinformatics and working with scRNA-seq data and libraries
• Experience in deep learning and NLP
• Experience in computational biology in any capacity

Helpful to have:

• Knowledge of wet-lab molecular biology
• Experience in fundamental algorithmic techniques

Prospective candidates should contact farhank@mit.edu

Master’s Project: Neural Implicit Fields for Representing Cardiovascular Organs

Project Description: In the realm of biomedical engineering and computational anatomy, the generation of synthetic 3D cardiovascular anatomy holds immense potential for advancing in-silico trials of device deployment. However, the complexity and volumetric nature of such data present substantial challenges for conventional 3D models. This master’s thesis project proposes to overcome these limitations by developing an implicit neural representation (INR) of 3D cardiovascular organs. INRs have demonstrated superior capabilities in capturing intricate shape details and offer greater scalability compared to traditional approaches. The core of this project involves not only the development and optimization of the INR framework but also its integration with existing generative models of cardiovascular anatomy. The student will benchmark the developed system against current state-of-the-art models to demonstrate its efficacy and advantages in the field.

Responsibilities: The candidate will undertake various critical responsibilities, including:

• Cultivating and managing diverse cardiovascular shape datasets.
• Developing and fine-tuning the INR framework to accurately represent 3D cardiovascular structures.
• Integrating the INR framework with generative models specifically designed for cardiovascular anatomy.

Prerequisites:

• Substantial experience in deep learning methodologies.
• Proficiency in computational geometry and mesh processing.
• A strong background in image processing.

Helpful to have:

• Expertise in the field of computer vision.
• Experience working with implicit neural representations.
• Knowledge and experience in shape analysis.

This project offers a unique opportunity to contribute to groundbreaking research in computational anatomy, with a specific focus on the development of advanced methods for 3D cardiovascular anatomy generation. The successful candidate will play a key role in pushing the boundaries of what is possible in this exciting and evolving field.

Prospective candidates should contact kkadry@mit.edu

Master’s Project: Interactive Multi-Tissue Segmentation Platform for Intravascular Imaging

Project Description: The development of digital twins of coronary arteries is a pivotal advancement in understanding and managing coronary artery disease. These digital representations, created from intravascular imaging segmentations such as optical coherence tomography (OCT), enable the simulation of various biophysical dynamics pertinent to the pathophysiology and intervention strategies of coronary artery disease. However, a critical challenge lies in the generalization of automated segmentation algorithms across different OCT systems and diverse coronary pathologies. This master’s thesis project aims to address this gap by developing an interactive multi-tissue segmentation platform specifically tailored for intravascular images. The platform will facilitate the rapid creation of high-quality segmentations for various arterial tissues, crucial for the creation of accurate 3D digital twins for biophysical simulations. Moreover, the project emphasizes the efficiency of the segmentation process, ensuring both precision in segmentation and expeditious quality control.

Responsibilities: The candidate will be engaged in a series of tasks integral to the success of this project, including:

• Organizing and managing intravascular imaging datasets.
• Developing and refining interactive segmentation models, along with adapting foundational models for specific requirements.
• Benchmarking the segmentation process, focusing on both the time efficiency and quality of the segmentations obtained.

Prerequisites:

• Proficiency in deep learning frameworks.
• Strong software engineering skills.
• A comprehensive understanding of image processing techniques.

Helpful to have:

• Familiarity with medical imaging modalities.
• Experience with 3D segmentation software.

This project presents a unique opportunity to contribute to the field of biomedical engineering, particularly in enhancing the understanding and treatment of coronary artery disease through advanced computational techniques. The successful candidate will play a pivotal role in advancing this innovative and impactful research.

Prospective candidates should contact kkadry@mit.edu

Master’s Project: Topological Regularization for Generative Models of Cardiovascular Anatomy

Project Description: In the realm of biomedical engineering and computational anatomy, the generation of synthetic 3D cardiovascular anatomy holds immense potential for advancing in-silico trials of device deployment. A key aspect of ensuring the utility of these models lies in their ability to generate topologically accurate representations. Accurate topology in this context means not only having the correct number of topological components such as connected components, loops, and voids, but also ensuring proper topological interactions between different tissue components. These interactions include containment, adjacency, and exclusion. This is crucial for enabling accurate numerical simulation of cardiovascular physics, as even minor topological defects can lead to unphysiological physical effects. This thesis project focuses on the development and implementation of topological regularization methods designed to penalize such defects in the output of generative models, thereby ensuring the generation of topologically viable anatomical configurations.

Responsibilities: The student will be responsible for several critical aspects of this project, including:

• Development of a framework for the detection and quantification of topological defects in cardiovascular anatomy.
• Development of deep-learning-based approaches for regularizing the topological quality of these models.
• Benchmarking the effectiveness of these approaches against existing state-of-the-art algorithms in the field.

Prerequisites:

• Knowledge and experience in computational geometry.
• Proficiency in deep learning methodologies.
• A strong background in image processing.

Helpful to have:

• Experience in persistent homology.
• Experience in skeletonization techniques.
• Familiarity with medical imaging technologies and data.

This project represents a significant step forward in the field of biomedical engineering, particularly in enhancing the accuracy and utility of synthetic models for cardiovascular anatomy. The successful candidate will contribute to this innovative research area, with the potential to influence the future of in-silico trials and medical imaging.

Prospective candidates should contact kkadry@mit.edu