📚 Daily Literature Update

Summary: This paper introduces READ, a retrieval-alignment diffusion framework for structure-based drug design (SBDD). It focuses on generating molecular ligands that bind to protein targets at atomic resolution, directly connecting machine learning-based generative modeling with protein-ligand interaction prediction.

Summary: This paper develops a geometric-chemical distance metric for comparing protein surfaces, where geometry and chemical patterning determine molecular interaction and recognition. It establishes both a distance measure and a correspondence between complete protein surfaces, which is fundamental for understanding how proteins bind and interact.

Summary: This study systematically evaluates attention mechanisms in protein language models for predicting functions of human olfactory receptors, linking attention weights to biologically relevant key regions. It directly addresses interpretability of protein language models, a core interest of the researcher.

Summary: This paper benchmarks quantization of ESM-2 protein language models for protein variant-effect prediction, showing that average metrics often hide failure cases in deep mutational scanning tasks. The work is directly relevant to the researcher's use of protein language models and predictive modeling of molecular effects.

Summary: This paper presents a classification framework for intracellular protein patterns in multi-component reaction-diffusion networks, directly relevant to systems biology and protein interaction dynamics. It addresses the complexity of locating instabilities in biological pattern-forming systems, which connects to computational approaches for understanding biomolecular organization and signaling.

Summary: This paper presents OCOO-T, a simple and scalable virtual cell model for predicting single-cell transcriptional responses to genetic, chemical, and cytokine perturbations. It contributes to AI Virtual Cell (AIVC) modeling with direct implications for drug discovery and understanding of cellular systems.

Summary: This paper introduces Archimap, a spatial transcriptomics method that enables in situ discovery of highly diverse and low-abundance immune repertoire sequences and therapeutic antibodies. It directly addresses biomolecular sequence analysis and interaction discovery, leveraging computational approaches that align with machine learning and protein language model interests.

Summary: This paper addresses the evaluation of computational methods for predicting cryptic binding sites, a key problem in protein interaction and structure-based drug discovery. It argues that current top-n recovery metrics conflate two independent abilities—correctly proposing a binding site location and ranking it highly—and separates these for more informative benchmarking. The work directly informs the researcher's interest in biomolecular interactions and predictive modeling.

Summary: This paper applies 3D Zernike descriptors to cluster antibody structures and identify repurposing opportunities based on structural and physicochemical similarities. It directly relates to the researcher's interests in protein structure, molecular recognition, and machine learning approaches to biomolecular data. The method bridges structural biology and predictive modeling of antibody cross-reactivity.

Summary: This paper evaluates masked language models (MLMs) for context-aware codon optimization, a task central to nucleic acid-based medicine. The work directly overlaps with the researcher's use of language models on biological sequences, extending the approach from proteins to codons. It assesses whether MLM-based methods outperform traditional frequency-based optimization, providing insights applicable to protein language model design.

Summary: LEN-Seek introduces a latent-space similarity search for ligand binding sites using an SE(3)-invariant graph VAE, enabling fast and scalable retrieval of structurally similar sites. This work aligns with the researcher's interest in structure-based machine learning and molecular recognition.

Summary: This paper proposes a "process bigraphs" framework for compositional systems biology, enabling integration of independently developed biological submodels with shared variables and coordinated execution. It addresses multiscale modeling challenges in systems biology through formal graph-based composition.

Summary: This paper investigates gene prioritization via network centrality in biological interaction networks, a core topic in systems biology. It shows that degree-based rankings systematically omit cross-module connector genes, and introduces a partition-free centrality measure that recovers them. The findings are relevant to the researcher's work on protein interaction networks and pattern discovery.

Summary: This paper applies sparse autoencoders to interpret the hidden states of BiRNA-BERT, an RNA language model, revealing structural and family-level features. Though focused on RNA, the methodology directly aligns with the researcher's interest in using language models for biomolecular sequence interpretation and pattern discovery.

Summary: SAMP V2 presents a stacking ensemble learning model for antimicrobial peptide (AMP) identification using augmented split amino acid composition and biochemical sequence-order information. The work leverages sequence-based machine learning to predict peptide function, relevant to the researcher's focus on biomolecular sequence pattern discovery.

Summary: This paper introduces a novel approach for automated inference of graph transformation rules, applicable to dynamic systems modeling in the life sciences. Graph transformation provides an expressive formalism for representing and reasoning about biomolecular interaction networks and dynamic cellular processes.