Seminars

August 25, 2025: (Special Seminar – 4PM)- Alexander Stark, Research Institute of Molecular Pathology
Decoding Transcriptional Regulation.
Host: Li Zhao
In higher eukaryotes, genes are expressed dynamically in complex spatial and temporal patterns, which are progressively refined to set up body plans and define specific cell-types. The information about when and where each gene is to be expressed is encoded in the sequences of promoter-, enhancer- and silencer regions and realized by transcription factor and cofactor proteins. I am presenting our work towards understanding the how this regulatory information is sequence-encoded and how cells utilize this information with transcription factor and cofactor proteins. We characterize regulatory sequences by functional screens in cell lines and by a genome-scale candidate testing approach in developing Drosophila embryos. We then use deep-learning approaches to model the sequence-to-function relationship for enhancers and build synthetic enhancers de novo. We also employ functional screens, mutagenesis, and biochemistry to dissect functions and mechanisms of transcription regulation in flies and mammals.
September 16, 2025: (4PM)- Hava Siegelmann, University of Massachusetts, Amherst
AI for Autonomous Agents: Sequence AI and Peer Cooperative Lifelong Learning.
Host: Marcelo Magnasco
How come drones are still mainly human controlled and have such limited autonomy? First, drones operate under significant constraints, including limited computational power, energy capacity, and communication bandwidth. Reinforcement Learning fail to maintain optimal performance under such constraints. We propose sequence AI algorithms that significantly improving compute and energy efficiency. Among the key features are rapid onboard responses and adaptability in dynamic environmental changes, robustness to missing inputs, minimization of sensor usage and the ability to use cheaper sensors to greater effect, as well as making possible the use of cheaper hardware while maintaining peak effectiveness. Second issue is the need of communication and cooperation among drones. Distributed AI is known to suffer explosion of communication needs, and this is not available in realistic swarms of drones. We propose a cooperative AI where the agents are lifelong learners. On the go, they are able to update, learn from failures, and become more expert with more experience. This paradigm enables both collaborative AI without explosive communication as well as a great reduction in the required labeled data (teacher), since the agents peer-teach each other. We suggest that these two directions of research will advance us towards true safe autonomy.
September 30, 2025: (4PM)- Suckjoon Jun, University of California, San Diego
Decoupling Of Global Metabolic Flux And Proteome Partitioning In Bacteria.
Host: Avi Flamholz
Bacteria regulate homeostatic growth by adjusting proteome composition. In Escherichia coli, this coordination is mediated by (p)ppGpp, which couples amino acid supply with ribosome production. We identify a distinct architecture in Bacillus subtilis, where GTP — not (p)ppGpp — controls proteome allocation. Translational inhibition results in GTP depletion and suppresses amino acid biosynthesis via feedback inhibition without altering ribosome abundance, decoupling global amino-acid flux from proteome composition. By artificially tuning GTP, we recoupled flux and proteome, restoring growth to maximal levels. The regulated sub-optimality enables a tradeoff to balance growth and stress resilience. Similar GTP-based strategies suggest evolutionary conservation in Firmicutes. These findings call for a revised view of bacterial physiology, where proteome composition and metabolic flux represent distinct regulatory layers.
October 14, 2025: (4PM)- Alex Williams, New York University
Quantifying Individuality In Neural Circuit Representations.
Host: Nikolas Schonsheck
Signatures of neural computation are thought to be reflected in the coordinated activity of large neural populations. Neuroscience is now flush with measurements of these activity patterns in humans, animal subjects, and large-scale artificial network models. In this talk, I will address an extensively studied, yet unresolved, question: How should we quantify the extent to which two or more neural circuits have “similar” activation patterns? Without an answer to this question, the field has struggled to investigate basic questions about biological variability and individuality, such as: How do neural representations vary across a healthy population? How do differences in neural population activity correlate with behavioral idiosyncrasies and disorders? How similar are computational mechanisms in biological brains and artificial neural networks? In this talk, I will summarize several mathematical methods that quantify similarity in neural representations and demonstrate how they provide early insights into these questions when applied to biological data and artificial networks.
October 21, 2025: (4PM)- Timothy P. Jenkins, Technical University Of Denmark
Translational Protein Design: How AI Is Transforming Therapeutic Discovery In Snakebite And Cancer.
Host: Jialong Jiang
Advances in generative AI are transforming how we discover and engineer therapeutic proteins. Rather than relying on random screening or animal immunisation, these models can now learn the fundamental principles of molecular recognition and design proteins capable of binding virtually any target. This talk will illustrate how such methods are being applied to address two very different biomedical challenges. The first involves designing synthetic proteins that neutralise deadly snake venoms by targeting key neurotoxins and cytotoxins. These de novo binders were created using AI-driven structural generation and achieved potent neutralisation both in vitro and in vivo, offering a potential path to safer, scalable, and globally accessible antivenom therapies. The second example focuses on targeting peptide–MHC complexes that present intracellular cancer antigens on the cell surface. Using generative models, we designed minibinders that specifically recognise these complexes and, when incorporated into immune cell receptors, enable precise killing of tumour cells. Together, these studies demonstrate how AI-guided protein design can move beyond prediction to invention, unlocking new therapeutic strategies across neglected diseases and advanced immunotherapies alike.
October 28, 2025: (4PM)- Frederick A. Matsen, Fred Hutchinson Cancer Research Center
Inviting Darwin Into Antibody Language Models.
Host: Gabriel Victora
Antibodies are coded by nucleotide sequences that are generated by V(D)J recombination and evolve according to nucleotide mutation and selection processes. Existing antibody language models, however, focus exclusively on antibodies as strings of amino acids and are fit using the masked language modeling objective. In this talk, I will first show that fitting using this objective implicitly incorporates nucleotide-level processes as part of the protein language model, which degrades performance when predicting functional properties of antibodies. To address this limitation, we propose a new framework: a deep amino acid selection model (DASM) that predicts the selective effect of replacing every amino acid with every alternate amino acid. By fitting selection as a separate term from the mutation process, the DASM exclusively quantifies functional effects. This separation of concerns leads to substantially improved performance on standard functional benchmarks. Moreover, our model is an order of magnitude smaller and orders of magnitude faster to evaluate than existing approaches, as well as being readily interpretable. I will then describe some surprising conclusions about how natural selection works for antibodies: there is more to the story than framework vs CDRs!
November 4, 2025: (4PM)- Manish Saggar, Stanford University
Beyond The Mean: Revealing The Hidden Scaffolding Of Brain Dynamics Through Topological Data Analysis.
Host: Nikolas Schonsheck
Understanding how the brain dynamically coordinates cognition, emotion, and behavior remains a central challenge in neuroscience and psychiatry. Traditional neuroimaging approaches often rely on static or averaged representations of brain activity, which obscure the rich temporal structure and inter-network transitions that may underlie both healthy and pathological states. In this talk, I will introduce a novel framework that combines topological data analysis (TDA) with graph-theoretic modeling to uncover reproducible and interpretable patterns in brain dynamics. Using the Mapper algorithm, we construct individual-level brain-state graphs that reveal flexible or rigid transitions between functional states across resting-state and task-based fMRI. These dynamic motifs not only reflect traits such as rumination and attentional flexibility but also offer new biomarkers for clinical populations. I will also present our recent efforts to integrate deep learning and latent factor modeling with the Research Domain Criteria (RDoC) framework. This work provides a data-driven refinement of mental health constructs by identifying latent “motifs” that better explain individual differences in brain activation across multiple tasks. Together, these approaches highlight the importance of moving beyond summary statistics toward manifold-level insights into brain function, with implications for both precision psychiatry and the development of AI-driven diagnostic tools.
November 11, 2025: (4PM)- Yufeng Shen, Columbia University
Genetics Of Developmental Disorders And Computational Models Of Missense Variants.
Host: Li Zhao
Developmental disorders, including congenital anomalies and neurodevelopmental disorders (NDD), affect many newborns and children. They are major factors associated with mortality, mobility, and long-term health of children and surviving adults. Finding genetic causes is fundamentally important for improving health care of these individuals and advancing basic research in human developmental biology. Previous efforts, which largely focused on de novo variants and loss of function variants with large effect, discovered new risk genes and pathways and were able to identify genetic causes in about 20-40% of “complex” cases who have additional congenital anomalies or NDD. The same approach was much less successful for “isolated” cases who do not have additional anomalies or NDD. A major analytical gap in previous studies is about rare inherited variants, especially missense variants, due to uncertainty of their genetic consequence and overall low signal-to-noise ratio. We reason that a key to address this issue is to leverage the fact that early onset severe conditions are under strong negative selection, therefore, genetic causes with moderate to large effect are under discernable negative selection. In this talk, I will go through MisFit, a new machine method developed based on population genetics models for estimating selection coefficient of missense variants. I’ll then show new results in autism regarding rare inherited variants and implications for rare congenital anomalies.
November 18, 2025: (4PM)- Elias Barriga, Technical University Dresden
Mechanoelectrical Actuation Of Tissue Morphogenesis.
Host: Amy Shyer/Alan Rodrigues
Cell state transitions are essential for embryos to develop, wounds to heal and diseases to progress. Still, the mechanisms that coordinate the transitions of cells within tissues from one state to another are not fully understood. In my talk I will discuss our advances in studying how physical forces that emerge from tissue interplay integrate with biochemical frameworks to coordinate robust cell state transitions in time and space.
December 2, 2025: (4PM)- Mariela Petkova, Harvard University
Exploring The Tension Between Fidelity And Variability In Biology From Genetic To Neural Networks.
Host: Avi Flamholz
Biological systems constantly navigate a delicate balance between reproducibility and variability. Developmental processes exemplify remarkable precision—each person has five fingers —but neural circuits in the brain must maintain a flexible architecture to enable adaptive behavior. In this talk, I explore this tension through two complementary lenses: small genetic networks guiding precise cell identities in fly development, and large neural networks guiding animal behavior. Specific examples include how electric fish subtract self-generated electrical signals to accurately detect prey, and how zebrafish swim upstream in complete darkness by integrating local water flow rotations in a neural implementation of Stoke’s theorem. By performing quantitative measurements of gene expression and neural connectivity and relating them to the biological function of each network, I demonstrate that simple rules can emerge even from the most intricate networks.
December 9, 2025: (4PM)
To Be Announced.
Host: TBD
To come.
December 16, 2025: (4PM)- Mohammed AlQuraishi, Columbia University
Some Observations On How AlphaFold Predicts, And Learns To Predict, Protein Structures.
Host: Jialong Jiang
In this talk I will discuss some recent evidence regarding the degree to which AlphaFold appears to learn to do implicit physics, as well as how this knowledge is acquired during the training process. Time permitting, I will also discuss differences between AlphaFold 2 and 3 with regards to the question of implicit physical knowledge.
January 6, 2026: (4PM)- Deepak Krishnamurthy, University of California, Berkeley
Microscale Biophysics of the Ocean: A Single-Cell Perspective on the Biological Carbon Pump.
Host: Avi Flamholz
Our oceans make up 90% of the planet’s biosphere. Yet, paradoxically, this vast ecosystem is built from the bottom up, with single cells driving its most critical processes. Microbial phytoplankton alone generate 50% of Earth’s oxygen and fix half of all carbon, matching the impact of terrestrial life. In the sunlit upper ocean, single-cell biomass and planktonic products aggregate and sink as “marine snow,” powering the “biological carbon pump,” one of Earth’s largest carbon fluxes estimated at 5–12 GtC per year. Mechanistically unraveling this pump demands connecting microscale cellular processes to ecological-scale phenomena—a challenge requiring new biophysical approaches.
In the first part of my talk, I’ll present our development of scale-free vertical tracking microscopy—a new system for measuring single cells, marine larvae, and sinking aggregates with microscopic resolution across ecologically relevant depths of tens to hundreds of meters. I’ll share vignettes of discoveries enabled by this tool, including novel cell motility in diatoms, behavioral repertoires of key marine larvae in “virtual-reality” environments, and the impact of single-cell processes on flow and mass transport to sinking aggregates.
In the second part, I’ll discuss ongoing work using choanoflagellates as a model for the biophysics of cell adhesion and behavior in the ocean. Protists—a remarkably diverse class of unicellular eukaryotes, including choanoflagellates—constitute 2 GtC of marine biomass, equal to all marine animals. Yet their ecosystem roles are just starting to be uncovered. I focus on loricate choanoflagellates: single cells that rapidly (in minutes) build intricate, silicaceous “lorica” baskets from hundreds of rod-like costal strips (2–3 µm long, 100 nm wide). I’ll explore how these cells modulate surface adhesion interactions in space and time, emphasizing the role of glycans, charge, and biophysical surface properties. Loricate choanoflagellates offer a powerful model to probe the limits of single cell behavior, microscale biological self-assembly, and modulation of adhesive interactions in the marine environment.
I’ll close with future directions: developing high-throughput assays at both molecular and cellular scales to measure adhesion interactions between cells and between cells and surfaces, with applications to marine snow formation and its emergent biophysical properties. These approaches are critical for biophysically constraining carbon flux models, elucidating the nonlinear microscale interactions that cause flux attenuation with depth, and predicting how the biological carbon pump will respond to anthropogenic stressors.
January 13, 2026: (4PM)
To Be Announced.
Host: TBD
To come.
January 20, 2026: (4PM)- Elad Schneidman, Weizmann Institute of Science
To Be Announced.
Host: Merav Stern
To come.
January 27, 2026: (4PM)- Enrique Rojas, New York University
To Be Announced.
Host: Avi Flamholz
To come.
February 3, 2026: (4PM)- Ashok Litwin-Kumar, Columbia University
To Be Announced.
Host: Nikolas Schonsheck
To come.
February 17, 2026: (4PM)- Noah Mitchell, University of Chicago
To Be Announced.
Host: Jialong Jiang/Dillon Cislo
To come.
February 24, 2026: (4PM)- Daniel Needleman, Harvard University (Location: Smith Hall Annex, A-Level, Physics Seminar Room)
To Be Announced.
Host: Avi Flamholz
To come.
March 10, 2026: (4PM)- Mo Ebrahimkhani, University of Pittsburgh
To Be Announced.
Host: Alan Rodrigues
To come.
March 31, 2026: (4PM)- Andreas Tolias, Stanford University
To Be Announced.
Host: Nikolas Schonsheck
To come.
April 7, 2026: (4PM)- Jonathan Pillow, Princeton University
To Be Announced.
Host: Nikolas Schonsheck
To come.
April 14, 2026: (4PM)- Eric Shea-Brown, University of Washington
To Be Announced.
Host: Merav Stern
To come.
April 21, 2026: (4PM)- Jennifer Schwarz, Syracuse University
To Be Announced.
Host: Nikolas Schonsheck
To come.
April 28, 2026: (4PM)- Johnatan (Yonatan) Aljadeff, University of California, San Diego
To Be Announced.
Host: Merav Stern
To come.
May 5, 2026: (4PM)- Yun S. Song, University of California, Berkeley
To Be Announced.
Host: Jialong Jiang
To come.
May 12, 2026: (4PM)- Arseny Finkelstein, Tel Aviv University
To Be Announced.
Host: Merav Stern
To come.