Transcription in individual cells is an inherently stochastic process, with non-deterministic numbers of RNA molecules produced per gene in transcriptional events across cells of the same cell type and condition. These distributions can be mathematical modeled according to biophysical hypotheses that specify system parameters of interest, such as rates of RNA splicing and degradation. While such stochastic models have been applied to infer RNA processing rates in fluorescence based assays, the scale of high-throughput sequencing data makes inference in this setting computationally challenging, especially as more molecular species, such as unspliced and spliced RNA molecules, can be quantified. We adapt an inference tool from machine learning, the variational autoencoder, that allows summarization and analysis of high-dimensional datasets to work with biophysical models of transcriptional dynamics. While preserving the variational autoencoder’s ability to reduce dimensionality of and denoise data, this work presents two major advances: one, principled integration and treatment of multimodal data (i.e., unspliced and spliced RNA molecules) through the employment of specific biophysical models; and two, inference of biophysical rates that parameterize those models, allowing the exploration of RNA dynamics across thousands of individual genes and cells.
Speaker: Maria Carilli
Affiliation: California Institute of Technology
Position: PhD Student, Lior Pachter Lab
Host: Dan MacDonald, PhD, Gibson Lab
Date: Monday April 28th, 2025
Time: 4:00-5:00PM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866
Relevant Papers: Biophysical modeling with variational autoencoders for bimodal, single-cell RNA sequencing data, Spectral neural approximations for models of transcriptional dynamics
Maria Carilli is a graduate student at Caltech working on theory and software development for the analysis of single-cell RNA sequencing data. She obtained undergraduate degrees in biophysics and music from the University of Colorado Boulder with a minor in computational biology. She is particularly interested in combining tools from machine learning with mathematical modeling of biophysical dynamics for deeper, and scalable, understanding of cellular systems.
Click here to be added to our mail list.
Medical imaging plays a critical role in modern healthcare, yet its complexity necessitates advanced computational tools to enhance analysis, interpretability, and diagnostic accuracy. In this presentation, we will discuss our research efforts in developing computational imaging biomarkers and frameworks for precision medicine in real clinical scenarios involving imperfect data. We will cover a spectrum of computational techniques grounded in both biological and domain-specific insights that facilitate the early detection and evaluation of treatment responses across various diseases.
In this talk, I aim to present insights into the design of evaluations of machine learning and AI systems to assess properties related to algorithmic fairness and health equity. I argue that such evaluations are contextual and require specification of the intended use, target population, and assumptions regarding the data generating process and measurement mechanism. Through examples from my research, I argue that causal graphical models can serve as key tools for context specification and can aid in understanding and appropriate use of analytical algorithmic fairness techniques. I will further discuss recent work related to evaluation of equity-related biases in large language models.
The genome is a sequence that encodes the DNA, RNA, and proteins orchestrating an organism’s function. We present Evo, a long-context genomic foundation model with a frontier architecture trained on millions of prokaryotic and phage genomes, and report the first scaling laws on DNA to complement observations in language and vision. Evo generalizes across DNA, RNA, and proteins, enabling zero-shot function prediction competitive with domain-specific language models and the generation of functional CRISPR-Cas and transposon systems, representing the first examples of protein-RNA and protein-DNA co-design with a language model. Evo also learns how small mutations affect whole-organism fitness and generates megabase-scale sequences with plausible genomic architecture. These prediction and generation capabilities span molecular to genome scales of complexity, advancing our understanding and control of biology.
The integration of Artificial Intelligence (AI) in clinical pathology has faced significant hurdles due to constraints in data collection and challenges associated with model transparency and interpretability. In this talk, we introduce a novel digital pathology AI framework named nuclei.io, which leverages active learning and incorporates real-time feedback from human experts. This innovative approach empowers pathologists to quickly generate diverse datasets and develop models for various clinical applications. To demonstrate the effectiveness of our framework, we conducted two user studies employing such human–AI collaboration strategy. These studies focused on two key areas: the identification of plasma cells in endometrial biopsies and the detection of colorectal cancer metastasis in lymph nodes. The results from these studies showed significant enhancements in sensitivity, accuracy, and diagnostic efficiency with the integration of AI. Our findings underscore the benefits of the human-in-the-loop AI framework, highlighting its potential to transform the field of digital pathology.
Advancements in single-cell and spatial-omics technologies have created a need for integrating datasets across modalities with limited and weakly correlated features, such as those between spatial proteomics and transcriptomics. Existing tools, usually designed for strongly linked data, often fail in these scenarios. Recently, we have developed a series of methods (Mario and MaxFuse), that improves integration by refining weak correlations between modalities through an iterative smoothing and co-embedding process, and achieves single-cell level matching across these weakly linked modalities, enabling in-depth understanding of tissue micro-environments.
BWH Computational Pathology Special Seminar
Human tissue, which is inherently three-dimensional (3D), is traditionally examined through standard-of-care histopathology as limited two-dimensional (2D) cross sections that can poorly represent the tissue due to sampling bias. To holistically characterize 3D histomorphology, 3D imaging modalities have been developed, but clinical translation is hampered by the complex and time-consuming requirements for manual evaluation, as well as the current lack of computational platforms to distill clinical insights from these large, high-resolution datasets. We present a deep learning model for processing tissue volumes and predicting patient outcomes with weak supervision. Recurrence risk-stratification models were trained with archived prostate cancer specimens imaged with open-top light-sheet microscopy or microcomputed tomography. By comprehensively capturing 3D morphologies, 3D block-based prognostication achieves superior performance to traditional 2D slice-based approaches, including existing clinical/histopathological baselines. Incorporating larger tissue volumes is shown to improve prognostic accuracy. This framework offers a promising direction for clinical decision support and 3D biomarker discovery, with the potential to further catalyze the growth of 3D spatial biology techniques for clinical applications.
Cell identity drives cell-cell communication and tissue architecture and is in return regulated by cell-extrinsic cues. Cell identity is determined by the combination of intrinsic developmentally established transcription factor use (TF) and constitutive as well as cell communication-dependent TF activities. Presented work shows two probabilistic models that we developed to advance the understanding of these processes using single-cell and spatial genomic data.