Biology subjects

Hofmann, J. L.

Publications and source records attributed to Hofmann, J. L..

5 recordsLinked to original sources

bioRxiv2025

Lab-in-the-loop therapeutic antibody design with deep learning

Therapeutic antibody design is a complex multi-property optimization problem with substantial promise for improvement with the application of machine-learning methods. Towards realizing that promise, we introduce "Lab-in-the-loop," a new approach that orchestrates state-of-the-art repertoire mining methods, generative machine learning models, multi-task property predictors, active learning ranking and selection, and in vitro experimentation in a semi-autonomous, iterative optimization loop. By automating the design of antibody variants, property prediction, ranking and selection of designs to assay in the lab, and ingestion of in vitro data, we enable an end-to-end approach to developing computationally-informed therapeutic antibody design pipelines. We apply lab-in-the-loop to eleven seed antibodies obtained via animal immunization with four clinically relevant antigen targets: EGFR, IL-6, HER2, and OSM. Over 1,800 unique antibody variants are tested throughout four rounds of iterative optimization identifying 3-100x better binding variants for all targets and 10/11 seeds, with the best binders exceeding 100 pM affinity, demonstrating a process by which end-to-end machine learning can be developed for therapeutic antibody development.

bioengineering↗

bioRxiv2025

DyAb: sequence-based antibody design and property prediction in a low-data regime

Protein therapeutic design and property prediction are frequently hampered by data scarcity. Here we propose a new model, DyAb, that addresses these issues by leveraging a pair-wise representation to predict differences in protein properties, rather than absolute values. DyAb is built on top of a pre-trained protein language model and achieves a Spearman rank correlation of up to 0.85 on binding affinity prediction across molecules targeting three different antigens (EGFR, IL-6, and an internal target), given as few as 100 training data. We employ DyAb in two design contexts: as a ranking model to score combinations of known mutations, and combined with a genetic algorithm to generate new sequences. Our method consistently generates novel antibody candidates with high binding rates, including designs that improve on the binding affinity of the lead molecule by more than ten-fold. DyAb represents a powerful tool for engineering therapeutic protein properties in low data regimes common in early-stage drug development.

bioengineering↗

bioRxiv2024

Macromolecular interactions and geometrical confinement determine the 3D diffusion of ribosome-sized particles in live Escherichia coli cells

The crowded bacterial cytoplasm is comprised of biomolecules that span several orders of magnitude in size and electrical charge. This complexity has been proposed as the source of the rich spatial organization and apparent anomalous diffusion of intracellular components, although this has not been tested directly. Here, we use biplane microscopy to track the 3D motion of self-assembled bacterial Genetically Encoded Multimeric nanoparticles (bGEMs) with tunable size (20 to 50 nm) and charge (-2160 to +1800 e) in live Escherichia coli cells. To probe intermolecular details at spatial and temporal resolutions beyond experimental limits, we also developed a colloidal whole-cell model that explicitly represents the size and charge of cytoplasmic macromolecules and the porous structure of the bacterial nucleoid. Combining these techniques, we show that bGEMs spatially segregate by size, with small 20-nm particles enriched inside the nucleoid, and larger and/or positively charged particles excluded from this region. Localization is driven by entropic and electrostatic forces arising from cytoplasmic polydispersity, nucleoid structure, geometrical confinement, and interactions with other biomolecules including ribosomes and DNA. We observe that at the timescales of traditional single molecule tracking experiments, motion appears sub-diffusive for all particle sizes and charges. However, using computer simulations with higher temporal resolution, we find that the apparent anomalous exponents are governed by the region of the cell in which bGEMs are located. Molecular motion does not display anomalous diffusion on short time scales and the apparent sub-diffusion arises from geometrical confinement within the nucleoid and by the cell boundary.

biophysics↗

bioRxiv2023

Pre-loading of translation molecules onto ribosomes speeds transport and protein synthesis in Escherichia coli

Translating ribosomes must wait after each elongation step for a new ternary complex (EF-Tu{middle dot}aa-tRNA{middle dot}GTP) to arrive, facilitating rapid codon recognition testing. We recently showed that this wait-time rate-limits elongation in Escherichia coli due to competitive combinatoric searching through crowded cytoplasm by thousands of E. colis 42 unique ternary complexes. Here, we investigate whether ribosomal L12 subunits pool translation molecules to reduce this wait time. We mimic transport and reactions underlying elongation in a physiologically accurate, physically-resolved model of crowded cytoplasm. We find that L12 pre-loading as much as doubles translation rate by reducing diffusive search time. But more L12 is not always better: faster-growing bacteria tend to have fewer L12. We resolve this apparent contradiction by demonstrating tradeoffs between binding and novel sampling as a function of copy number in E. coli. Variable L12 copy numbers may thus have evolved for fast or slow bacterial growth as complementary survival strategies.

biophysics↗

bioRxiv2022

Ultra-weak protein-protein interactions can modulate proteome-wide searching and binding

Research on protein-protein interaction (PPIs) tends to focus on high affinity interactions. Weaker interactions (Kd >1M) recently understood as contributing to intracellular phase separation suggest that even-weaker PPIs might also matter in as-yet unknown ways. However, ultra-weak PPIs (Kd >1mM) are not readily accessible by in vivo techniques. Here we use protein electrostatics to estimate PPI strengths and spatially-resolved dynamic simulations to investigate the potential impacts of ultra-weak PPIs within dense protein suspensions. We find that ultra-weak PPIs can drive formation of transient clusters that last long enough to enable enzyme-catalyzed reactions and accelerate the sampling of protein associations. We apply our method to Mycoplasma genitalium, finding that ultra-weak PPIs should be ubiquitous among cytoplasmic proteins. We also predict that the proteome-wide interactome can be shifted to favor binding-dominant ultra-weak PPIs via the introduction of a few charged protein complexes. We speculate that ultra-weak PPIs could contribute to cellular fitness by facilitating sampling and colloidal-scale transport of proteins involved in biological processes, including protein synthesis.

biophysics↗