Biology subjects

Miranda Quintana, R. A.

Publications and source records attributed to Miranda Quintana, R. A..

17 recordsLinked to original sources

bioRxiv2026

How effective are contrastive learning-based approaches for activity-cliff prediction?

Activity cliffs, defined as structurally similar molecules with vastly different properties represent a fundamental challenge in modern day drug discovery for property prediction models. While Graph Neural Networks (GNNs) have advanced molecular property prediction, they inherently struggle with this problem due to representation collapse and node over smoothening. In this work, we evaluate the efficacy of incorporating various contrastive learning-based loss functions, including the Supervised Contrastive (SupCon) configurations into several GNN backbones to navigate this structure-activity landscape. Tested across the MoleculeACE benchmarks, the results indicate that while contrastive losses prevent GNN mode collapse to some extent and drastically improve the property-based separation in the embedding space, their impact on absolute predictive accuracy remains highly dataset dependent. Ultimately, these methods offer critical gains in interpretability for mapping optimized lead generations, opening up new directions for models that respect these sharp pharmacological discontinuities.

bioinformatics↗

bioRxiv2026

Pathway Representation via Intrinsic Structural Medoids (PRISM): A Structural Mapping Approach to Clustering Molecular Pathways

We present Pathway Representation via Intrinsic Structural Medoids (PRISM), a state-aware framework for clustering pathways from molecular dynamics simulations of biomolecular transitions. In PRISM, each pathway is mapped to a small set of structural medoids obtained via a deterministic k-means clustering scheme. Pairwise pathway dissimilarities are computed using a weighted average Hausdorff distance between these representative sets, effectively capturing mean nearest-neighbor structural deviations while reducing sensitivity to outliers. Hierarchical agglomerative clustering of the resulting dissimilarity matrix defines pathway families. We evaluate PRISM across three biomolecular transitions of increasing complexity: alanine dipeptide C7eq [->] C7ax isomerization, adenylate kinase opening, and HIF-2 PAS-B ligand unbinding. PRISM consistently yields robust cluster assignments, with medoids faithfully representing distinct conformational states. By combining a state-based description with robust geometric dissimilarities, PRISM provides a scalable framework for organizing complex transition pathways.

biophysics↗

bioRxiv2026

Efficient exploration of peptide libraries using active learning with AlphaFold-based screening

We previously showed that AlphaFold2 can be used to screen for peptide-binding epitopes targeting the extraterminal (ET) domain of Bromodomain and Extraterminal (BET) proteins from candidate protein partners identified in pull-down experiments. However, such approaches require large numbers of AlphaFold2 calculations, making exhaustive screening impractical for larger datasets, such as viral proteomes that may target the ET domain. In many cases, identifying a substantial fraction of binders--even without exhaustive coverage--would already provide valuable biological insight into these interaction networks. Here, we show that an active learning strategy based on Thompson sampling (TS) can efficiently explore peptide sequence space. Using a library derived from BRD3 pull-down experiments, TS recovers 50% of all binders using 15% of the queries required by exhaustive sampling (3.3 times improvement over random sampling). Moreover, TS consistently identifies experimentally known binding epitopes with substantially fewer queries. Because the approach relies only on binary labels, it is readily transferable to other protein-peptide systems where AF-based binding classification is applicable, as well as to peptide-property predictors for properties such as solubility or aggregation propensity.

bioinformatics↗

bioRxiv2025

Best practices to cluster large molecular libraries

BitBIRCH is a novel clustering algorithm that enables the analysis of extremely large molecular libraries; however, its performance can be hindered by an excessive number of singletons or the formation of disproportionately large clusters. Here, we present a data-driven strategy to identify optimal BitBIRCH parameters that mitigate these limitations. Using the ChEMBL34 library as a case study (with additional datasets reported in the Supporting Information), we show that similarity thresholds between three and four standard deviations above the global mean provide a balanced trade-off between cluster count and medoid similarity. These values are efficiently approximate with the iSIM and iSIM-sigma frameworks. For the branching factor, values as high as computationally feasible are recommended, as increasing it to 1024 substantially reduced the number of singletons. We further introduce an iterative re-clustering procedure wherein the similarity threshold can be adjusted to merge related subclusters and singletons from the initial clustering, providing user-defined control over the extent of cluster fusion. This work provides practical guidelines to enhance the robustness and usability of BitBIRCH for large-scale molecular clustering.

bioinformatics↗

bioRxiv2025

Selector: A General Python Library for Diverse Subset Selection

Selector is a free, open-source Python library for selecting diverse subsets from any dataset, making it a versatile tool across a wide range of application domains. Selector implements different subset sampling algorithms based on sample distance, similarity, and spatial partitioning, along with metrics to quantify subset diversity. It is flexible and integrates seamlessly with popular Python libraries like Scikit-Learn, demonstrating the interoperability of the implemented algorithms with data analysis workflows. Selector is an operating-system agnostic, accessible, and easily extensible package designed with modern software development practices, including version control, unit testing, and continuous integration. Interactive quick-start notebooks, which are also web-accessible, provide user-friendly tutorials for all skill levels, showcasing applications in computational chemistry, drug discovery, and chemical library design. Additionally, a web interface has been developed that allows users to easily upload datasets, configure sampling settings, and run subset selection algorithms, with no programming required. This paper serves as the official release note for the Selector package, offering a technical overview of its features, use cases, and development practices that ensure its quality and maintainability.

bioinformatics↗

bioRxiv2025

mdBIRCH for Fast, Scalable, Online Clustering of Molecular Dynamics Trajectories

We present mdBIRCH, an online clustering method that adapts the BIRCH CF-tree to molecular dynamics (MD) data by applying a merge test calibrated directly to RMSD. Each arriving frame is routed to the nearest leaf microcluster and merged only if the post-merge centroid-based spread, computed from the cluster feature (CF) summaries, remains within a user-supplied threshold{tau} . This enables incremental, memory-bounded operation without constructing pairwise distance matrices, with a physically interpretable parameter controlling structural granularity. We evaluate mdBIRCH on {beta}-heptapeptide and HP35 systems and propose two practical strategies to make the threshold selection easier: (a) RMSD-anchored runs that use controlled structural edits to define interpretable operating points and (b) a blind sweep that tracks how cluster counts, occupancies, and coverage evolve with the threshold. Across both systems, increasing this threshold produces predictable consolidation into fewer, higher-occupancy states, and broader RMSD-to-centroid distributions. We further analyze the effect of the CF-tree capacity through the branching factor, quantify sensitivity to data ordering, and compare dominant-state representatives against batch clustering workflows to contextualize the resulting partitions. Finally, because decisions rely only on cluster summaries, mdBIRCH scales near-linearly with the number of frames on standard CPU hardware, offering a practical combination of speed and interpretability for large-scale trajectory analysis.

biophysics↗

bioRxiv2025

Undersampling techniques for large datasets

DNA-Encoded Libraries allow for an efficient approach to synthesize and screen billions of small molecules against a target of interest. With more real-world binding data, this can improve training of machine learning models. However, one key challenge in DELs is the severe imbalances between the classes, in other words, there are much more inactive than active compounds against any given target. This can heavily skew the training process. In this study, we explore different undersampling strategies for the majority class. These different techniques are benchmarked against random selection and prototyped on two different DEL datasets with three different machine learning models. Overall, the max_sim strategy shows the best scores, and the general pipeline is implemented in the DELight package.

bioinformatics↗

bioRxiv2025

Efficiently finding activity cliffs

Activity cliffs remain a key challenge in computational drug design, defying even modern machine learning approaches. Here, we present a dual approach that allows either navigating a structure-activity landscape to quickly identify activity cliffs, or how avoid them and identify maximally smooth sectors of chemical space. Both methods are built upon the BitBIRCH clustering algorithm, so they are ideally suited to analyze very large compound libraries. The code is freely available at https://github.com/mqcomplab/BitBIRCH_AC.

bioinformatics↗

bioRxiv2025

Divide and Cluster: The DIVINE Framework for Deterministic Top-Down Analysis of Molecular Dynamics Trajectories

We present DIVIsive N-ary Ensembles (DIVINE), a deterministic, top-down clustering framework designed for molecular dynamics (MD) trajectories. DIVINE constructs a complete clustering hierarchy by recursively splitting clusters based on n-ary similarity principles, avoiding the need for O(N2) pairwise distance matrices. It supports multiple cluster selection criteria, including a weighted variance metric, and deterministic anchor initialization strategies such as NANI (N-ary Natural Initiation), ensuring reproducible and structurally meaningful partitions. Testing DIVINE up to a 305 s folding trajectory of the villin headpiece (HP35) revealed that it matched or exceeded the clustering quality of bisecting k-means while reducing runtime and eliminating stochastic variability. Its single-pass design enables efficient exploration of clustering resolutions without repeated executions. By combining scalability, interpretability, and determinism, DIVINE offers a robust and practical alternative to conventional MD clustering methods. DIVINE is publicly available as part of the MDANCE package: https://github.com/mqcomplab/MDANCE.

biophysics↗

bioRxiv2025

BitBIRCH Clustering Refinement Strategies

Chemical libraries are becoming not only increasingly bigger, but they are doing so at an accelerated pace. Keeping up with this explosion in chemical data demands more than just hardware upgrades, we need dramatically more efficient algorithms as well. We have been working in this direction, with the introduction of the iSIM framework, which uses n-ary similarity to speed up the processing of very large sets. Recently, we showed how to use this technique to cluster billions of molecules with unprecedented efficiency through the BitBIRCH algorithm. In this Application Note we present a package fully-dedicated to expanding on the BitBIRCH method, including multiple options that give the user appreciable control over the tree structure, while dramatically improving the quality of the final partitions. Remarkably, this is achieved without compromising the efficiency of the original method. We also present new post-processing tools that help dissect the clustering information, as well as ample examples showcasing the new functionalities. BitBIRCH is publicly available at: https://github.com/mqcomplab/bitbirch.

bioinformatics↗

bioRxiv2025

iCliff Taylor's version: Robust and Efficient Activity Cliff Determination

Activity cliffs represent an important challenge to tackle in cheminformatics and drug design. One of the most common indicators to quantify them is the SALI index. Here we expose mathematical limitations of SALIs formulation, the most evident: it is undefined in instances where the similarity between two molecules is one. We show how using a simple Taylors series can aid this main problem, yielding a defined expression that can capture the ranking information from the original SALI. The second issue to solve is the quadratic complexity of using SALI to describe the roughness of the activity landscape of a set. Here, we propose iCliff, an indicator that can quantify the roughness in linear complexity. For this, we leverage the iSIM framework to obtain the average similarity of the set and a rearrangement to obtain the average of the squared property differences. The calculations for 30 different AC-focused databases suggest that there is a strong correlation between iCliff and the average pairwise of SALIs pairwise Taylor Series. To further explore the individual effects of removing each molecule in the activity landscape, we propose complementary iCliff. With this tool, we were able to identify the molecules that have a high number of activity cliffs with the rest of the molecules in the set.

bioinformatics↗

bioRxiv2025

Is Tanimoto a metric?

No. However, here we show how to generate a metric consistent with the Tanimoto similarity. We also explore new properties of this index, and how it relates to other popular alternatives.

bioinformatics↗

bioRxiv2025

Growth vs. Diversity: A Time-Evolution Analysis of the Chemical Space

Chemical space is a core and theoretical concept in cheminformatics, and it also has practical applications in drug discovery and other research areas. Chemical space is frequently associated with the number of molecules in the universe (e.g., chemical universe). It is well known that the number of compounds (both synthesized and theoretical ones) is rapidly increasing. It would be obvious to affirm that the chemical space is expanding (as a proxy of growth). But is the chemical diversity of compound libraries growing? In this study, we tackle this question by assessing quantitatively the time evolution of chemical libraries in terms of the chemical diversity as measured with molecular fingerprints. To tackle this task, we employed innovative cheminformatics methods to assess the progress over time of the chemical diversity of compound libraries available in the public domain. Using the iSIM and the BitBIRCH clustering algorithm, we conclude that, based on the fingerprints used to represent the chemical structures, just an increasing number of molecules cannot be directly translated to diversity for the analyzed libraries. With these tools, we have identified what releases contributed to the diversity of the library and the zones it did.

bioinformatics↗

bioRxiv2025

SHINE: Deterministic Many-to-Many clustering of Molecular Pathways

State-of-the-art molecular dynamics (MD) simulation methods can generate diverse ensembles of pathways for complex biological processes. Analyzing these pathways using statistical mechanics tools demands identifying key states that contribute to both the dynamic and equilibrium properties of the system. This task becomes especially challenging when analyzing multiple MD simulations simultaneously, a common scenario in enhanced sampling techniques like the weighted ensemble strategy. Here, we present a new module of the MDANCE package designed to streamline the analysis of pathway ensembles. This module integrates n-ary similarity, cheminformatics-inspired tools, and hierarchical clustering to improve analysis efficiency. We present the theoretical foundation behind this approach, termed Sampling Hierarchical Intrinsic N-ary Ensembles (SHINE), and demonstrate its application to simulations of alanine dipeptide and adenylate kinase.

biophysics↗

bioRxiv2024

Efficient clustering of large molecular libraries

The widespread use of Machine Learning (ML) techniques in chemical applications has come with the pressing need to analyze extremely large molecular libraries. In particular, clustering remains one of the most common tools to dissect the chemical space. Unfortunately, most current approaches present unfavorable time and memory scaling, which makes them unsuitable to handle million- and billion-sized sets. Here, we propose to bypass these problems with a time- and memory-efficient clustering algorithm, BitBIRCH. This method uses a tree structure similar to the one found in the Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) algorithm to ensure O(N) time scaling. BitBIRCH leverages the instant similarity (iSIM) formalism to process binary fingerprints, allowing the use of Tanimoto similarity, and reducing memory requirements. Our tests show that BitBIRCH is already > 1,000 times faster than standard implementations of the Taylor-Butina clustering for libraries with 1,500,000 molecules. BitBIRCH increases efficiency without compromising the quality of the resulting clusters. We explore strategies to handle large sets, which we applied in the clustering of one billion molecules under 5 hours using a parallel/iterative BitBIRCH approximation.

bioinformatics↗

bioRxiv2024

k-Means NANI: an improved clustering algorithm for Molecular Dynamics simulations

One of the key challenges of k-means clustering is the seed selection or the initial centroid estimation since the clustering result depends heavily on this choice. Alternatives such as k-means++ have mitigated this limitation by estimating the centroids using an empirical probability distribution. However, with high-dimensional and complex datasets such as those obtained from molecular simulation, k-means++ fails to partition the data in an optimal manner. Furthermore, stochastic elements in all flavors of k-means++ will lead to a lack of reproducibility. K-means N-Ary Natural Initiation (NANI) is presented as an alternative to tackle this challenge by using efficient n-ary comparisons to both identify high-density regions in the data and select a diverse set of initial conformations. Centroids generated from NANI are not only representative of the data and different from one another, helping k-means to partition the data accurately, but also deterministic, providing consistent cluster populations across replicates. From peptide and protein folding molecular simulations, NANI was able to create compact and well-separated clusters as well as accurately find the metastable states that agree with the literature. NANI can cluster diverse datasets and be used as a standalone tool or as part of our MDANCE clustering package.

biophysics↗

bioRxiv2023

Extended Similarity Methods for Efficient Data Mining in Imaging Mass Spectrometry

Imaging mass spectrometry is a label-free imaging modality that allows for the spatial mapping of many compounds directly in tissues. In an imaging mass spectrometry experiment, a raster of the tissue surface produces a mass spectrum at each sampled x, y position, resulting in thousands of individual mass spectra, each comprising a pixel in the resulting ion images. However, efficient analysis of imaging mass spectrometry datasets can be challenging due to the hyperspectral characteristics of the data. Each spectrum contains several thousand unique compounds at discrete m/z values that result in unique ion images, which demands robust and efficient algorithms for searching, statistical analysis, and visualization. Some traditional post-processing techniques are fundamentally ill-equipped to dissect these types of data. For example, while principal component analysis (PCA) has long served as a useful tool for mining imaging mass spectrometry datasets to identify correlated analytes and biological regions of interest, the interpretation of the PCA scores and loadings can be non-trivial. The loadings often containing negative peaks in the PCA-derived pseudo-spectra, which are difficult to ascribe to underlying tissue biology. Herein, we have utilized extended similarity indices to streamline the interpretation of imaging mass spectrometry data. This novel workflow uses PCA as a pixel-selection method to parse out the most and least correlated pixels, which are then compared using the extended similarity indices. The extended similarity indices complement PCA by removing all non-physical artifacts and streamlining the interpretation of large volumes of IMS spectra simultaneously. The linear complexity, O(N), of these indices suggests that large imaging mass spectrometry datasets can be analyzed in a 1:1 scale of time and space with respect to the size of the input data. The extended similarity indices algorithmic workflow is exemplified here by identifying discrete biological regions of mouse brain tissue.

bioinformatics↗