Biology subjects

Majewski, M.

Publications and source records attributed to Majewski, M..

4 recordsLinked to original sources

bioRxiv2026

Agent-driven Model Development for RNA 3D Structure Prediction

Large language model (LLM) agents have shown promise in driving scientific discovery, but their effectiveness in complex, real-world biological problems remains underexplored. We ask whether a general-purpose LLM agent can drive semi-autonomous development of a model for a genuinely hard biological problem, RNA 3D structure prediction. We designed a development loop where, under a fixed budget and with human supervision, the agent iteratively proposed, implemented, trained, and evaluated model changes. Over 297 iterations, the model evolved from a randomly-initialised baseline to QuickFold, an 8.9M-parameter folding trunk that matches the strongest open-source baselines (RhoFold+, NuFold) on lDDT and TM-score within noise on a held-out test set at a fraction of their inference cost. We frame this less as a new predictor than as a case study in feedback-driven, agent-led model development.

bioinformatics↗

bioRxiv2026

AlphaFold Database expands to proteome-scale quaternary structures

Protein function is governed by molecular interactions, yet structural coverage of these interactions remains sparse. The AlphaFold Protein Structure Database (AFDB) transformed access to accurate monomeric protein structures at scale. Here, we expand the AFDB to quaternary structures by predicting 31M candidate homo- and heterodimeric protein complexes, compiled from 4,777 proteomes, including model- and global health organisms. We established confidence criteria via analysis of experimentally determined structures, resulting in 1.81M high-confidence predictions. These models enabled the discovery of emergent structures and topologies not present in monomeric predictions. Additionally, the top 1% of structural clusters accounted for [~]44% of all complexes, and [~]8.3% of clusters were conserved across multiple domains of life, pointing to a substantial fraction of ancient, universally retained assemblies. Structural search highlighted that high-confidence predictions are anchored in experimental multimer space, yet 31.3% extend beyond detectable PDB coverage. These freely accessible proteome-scale predictions facilitate functional and mechanistic hypothesis generation across biology.

synthetic biology↗

bioRxiv2025

Synthetic genomic dissection of enhancer context sensitivity and synergy

Noncoding disease and trait-associated genetic variation is frequently interpreted in the context of genomic regulatory elements such as DNase I hypersensitive sites (DHSs). But while most DHSs lie within a few kilobases of another DHS, regulatory elements are typically analyzed individually without accounting for their neighbors. We characterize multiple heterotypic DHS combinations from different critical mESC regulator loci, all delivered in a constant chromosomal context replacing the Sox2 Locus Control Region (LCR). We employ an optimized high-throughput multiplexed delivery pipeline enabling analysis of 213 distinct payloads in 641 mouse embryonic stem cell (mESC) clones. We identify widespread examples of context-dependent enhancers which have no activity on their own but can more than double the activity of a neighboring DHS. Enhancers exhibit synergy only with certain partners, and deliveries to the Igf2/H19 locus show that synergy is not constrained to a single genomic context. We further show that synergy between neighboring DHSs decays as a characteristic function of distance, with its influence extending up to 4 kilobases. We fine map this context dependency to the contribution of individual transcription factor recognition sequences. Our approach implicates the specific sequence and architectural features underpinning pervasive genomic context effects, and outlines a direction for modeling the functional impact of noncoding regulatory variation on common human traits and diseases.

genomics↗

bioRxiv2018

Are protein-ligand complexes robust structures?

The predominant view in structure-based drug design is that small-molecule ligands, once bound to their target structures, display a well-defined binding mode. While this is convenient from a design perspective, it ignores the fact that structural stability (robustness) is not necessary for thermodynamic stability (binding affinity). In fact, any potential benefit of a rigid binding mode will have to be balanced against the entropic penalty that it entails. Surprisingly, little is known about the causes, consequences and real degree of robustness of protein-ligand complexes. Here we investigate two diverse sets of structures, comprising 79 drug-like and 27 fragment ligands, respectively. We focus on hydrogen bond interactions (469 in total), as they have been described as essential for structural stability. We find that 75% of complexes are anchored by at least one robust hydrogen bond, the remaining 25% either form loose complexes or are constrained by other interactions types. The first type of complexes generally combine a single anchoring point with looser regions, thus balancing order and disorder. Completely constricted protein-ligand complexes are rare and seem to fulfil a functional necessity. Structural stability analysis reveals a hidden layer of complexity in protein-ligand complexes that should be considered in ligand design.

biophysics↗