arXiv subjects

Kevin Michalewicz

Publications and source records attributed to Kevin Michalewicz.

8 recordsLinked to original sources

arXiv2025

Machine learning approaches for interpretable antibody property prediction using structural data

Understanding the relationship between antibody sequence, structure and function is essential for the design of antibody-based therapeutics and research tools. Recently, machine learning (ML) models mostly based on the application of large language models to sequence information have been developed to predict antibody properties. Yet there are open directions to incorporate structural information, not only to enhance prediction but also to offer insights into the underlying molecular mechanisms. This chapter provides an overview of these approaches and describes two ML frameworks that integrate structural data (via graph representations) with neural networks to predict properties of antibodies: ANTIPASTI predicts binding affinity (a global property) whereas INFUSSE predicts residue flexibility (a local property). We survey the principles underpinning these models; the ways in which they encode structural knowledge; and the strategies that can be used to extract biologically relevant statistical signals that can help discover and disentangle molecular determinants of the properties of interest.

q-bio.QM↗

arXiv2025

Protein generation with embedding learning for motif diversification

A fundamental challenge in protein design is the trade-off between generating structural diversity while preserving motif biological function. Current state-of-the-art methods, such as partial diffusion in RFdiffusion, often fail to resolve this trade-off: small perturbations yield motifs nearly identical to the native structure, whereas larger perturbations violate the geometric constraints necessary for biological function. We introduce Protein Generation with Embedding Learning (PGEL), a general framework that learns high-dimensional embeddings encoding sequence and structural features of a target motif in the representation space of a diffusion model's frozen denoiser, and then enhances motif diversity by introducing controlled perturbations in the embedding space. PGEL is thus able to loosen geometric constraints while satisfying typical design metrics, leading to more diverse yet viable structures. We demonstrate PGEL on three representative cases: a monomer, a protein-protein interface, and a cancer-related transcription factor complex. In all cases, PGEL achieves greater structural diversity, better designability, and improved self-consistency, as compared to partial diffusion. Our results establish PGEL as a general strategy for embedding-driven protein generation allowing for systematic, viable diversification of functional motifs.

q-bio.QM↗

arXiv2025

Integrating protein sequence embeddings with structure via graph-based deep learning for single-residue property prediction

Understanding the intertwined contributions of amino acid sequence and spatial structure is essential to explain protein behaviour. Here, we introduce INFUSSE (Integrated Network Framework Unifying Structure and Sequence Embeddings), a deep learning framework for the prediction of single-residue properties that combines fine-tuning of sequence embeddings derived from a Large Language Model with the inclusion of graph-based representations of protein structures via a diffusive Graph Convolutional Network. To illustrate the benefits of jointly leveraging sequence and structure, we apply INFUSSE to the prediction of B-factors in antibodies, a residue property that reflects the local flexibility shaped by biochemical and structural constraints in these highly variable and dynamic proteins. Using a dataset of 1510 antibody and antibody-antigen complexes from the database SAbDab, we show that INFUSSE improves performance over current machine learning (ML) methods based on sequence or structure alone, and allows for the systematic disentanglement of sequence and structure contributions to the performance. Our results show that adding structural information via geometric graphs enhances predictions especially for intrinsically disordered regions, protein-protein interaction sites, and highly variable amino acid positions -- all key structural features for antibody function which are not well captured by purely sequence-based ML descriptions.

q-bio.QM↗

arXiv2024

The MAPS Algorithm: Fast model-agnostic and distribution-free prediction intervals for supervised learning

A fundamental problem in modern supervised learning is computing reliable conditional prediction intervals in high-dimensional settings: existing methods often rely on restrictive modelling assumptions, do not scale as predictor dimension increases, or only guarantee marginal (population-level) rather than conditional (individual-level) coverage. We introduce the $\textit{lifted predictive model}$ (LPM), a new conditional representation, and propose the MAPS (Model-Agnostic Prediction Sets) algorithm that produces distribution-free conditional prediction intervals and adapts to any trained predictive model. Our procedure is bootstrap-based, scales to high-dimensional inputs and accounts for heteroscedastic errors. We establish the theoretical properties of the LPM, connect prediction accuracy to interval length, and provide sufficient conditions for asymptotic conditional coverage. We evaluate the finite-sample performance of MAPS in a simulation study, and apply our method to simulation-based inference and image classification. In the former, MAPS provides the first approach for debiasing neural Bayes estimators and constructing valid confidence intervals for model parameters given the estimators, at any desired level. In the latter, it provides the first approach that accounts for uncertainty in model calibration and label prediction.

stat.ML↗

arXiv2024

ANTIPASTI: interpretable prediction of antibody binding affinity exploiting Normal Modes and Deep Learning

The high binding affinity of antibodies towards their cognate targets is key to eliciting effective immune responses, as well as to the use of antibodies as research and therapeutic tools. Here, we propose ANTIPASTI, a Convolutional Neural Network model that achieves state-of-the-art performance in the prediction of antibody binding affinity using as input a representation of antibody-antigen structures in terms of Normal Mode correlation maps derived from Elastic Network Models. This representation captures not only structural features but energetic patterns of local and global residue fluctuations. The learnt representations are interpretable: they reveal similarities of binding patterns among antibodies targeting the same antigen type, and can be used to quantify the importance of antibody regions contributing to binding affinity. Our results show the importance of the antigen imprint in the Normal Mode landscape, and the dominance of cooperative effects and long-range correlations between antibody regions to determine binding affinity.

q-bio.QM↗

arXiv2024

Image deconvolution and PSF reconstruction with STARRED: a wavelet-based two-channel method optimized for light-curve extraction

We present STARRED, a Point Spread Function (PSF) reconstruction, two-channel deconvolution, and light curve extraction method designed for high-precision photometric measurements in imaging time series. An improved resolution of the data is targeted rather than an infinite one, thereby minimizing deconvolution artifacts. In addition, STARRED performs a joint deconvolution of all available data, accounting for epoch-to-epoch variations of the PSF and decomposing the resulting deconvolved image into a point source and an extended source channel. The output is a deep sharp frame combining all data, and the photometry of all point sources in the field of view as a function of time. Of note, STARRED also provides exquisite PSF models for each data frame. We showcase three applications of STARRED in the context of the imminent LSST survey and of JWST imaging: i) the extraction of supernovae light curves and the scene representation of their host galaxy, ii) the extraction of lensed quasar light curves for time-delay cosmography, and iii) the measurement of the spectral energy distribution of globular clusters in the "Sparkler", a galaxy at redshift z=1.378 strongly lensed by the galaxy cluster SMACS J0723.3-7327. STARRED is implemented in JAX, leveraging automatic differentiation and GPU acceleration. This enables rapid processing of large time-domain datasets, positioning the method as a powerful tool for extracting light curves from the multitude of lensed or unlensed variable and transient objects in the Rubin-LSST data, even when blended with intervening objects.

astro-ph.IM↗

arXiv2023

STARRED: a two-channel deconvolution method with Starlet regularization

The spatial resolution of astronomical images is limited by atmospheric turbulence and diffraction in the telescope optics, resulting in blurred images. This makes it difficult to accurately measure the brightness of blended objects because the contributions from adjacent objects are mixed in a time-variable manner due to changes in the atmospheric conditions. However, this effect can be corrected by characterizing the Point Spread Function (PSF), which describes how a point source is blurred on a detector. This function can be estimated from the stars in the field of view, which provides a natural sampling of the PSF across the entire field of view. Once the PSF is estimated, it can be removed from the data through the so-called deconvolution process, leading to images of improved spatial resolution. The deconvolution operation is an ill-posed inverse problem due to noise and pixelization of the data. To solve this problem, regularization is necessary to guarantee the robustness of the solution. Regularization can take the form of a sparse prior, meaning that the recovered solution can be represented with only a few basis eigenvectors. STARRED is a Python package developed in the context of the COSMOGRAIL collaboration and applies to a vast variety of astronomical problems. It proposes to use an isotropic wavelet basis, called Starlets, to regularize the solution of the deconvolution problem. This family of wavelets has been shown to be well-suited to represent astronomical objects. STARRED provides two modules to first reconstruct the PSF, and then perform the deconvolution. It is based on two key concepts: i) the image is reconstructed in two separate channels, one for the point sources and one for the extended sources, and ii) the code relies on the deliberate choice of not completely removing the effect of the PSF, but rather bringing the image to a higher resolution.

astro-ph.IM↗

arXiv2022

Deep Learning-based galaxy image deconvolution

With the onset of large-scale astronomical surveys capturing millions of images, there is an increasing need to develop fast and accurate deconvolution algorithms that generalize well to different images. A powerful and accessible deconvolution method would allow for the reconstruction of a cleaner estimation of the sky. The deconvolved images would be helpful to perform photometric measurements to help make progress in the fields of galaxy formation and evolution. We propose a new deconvolution method based on the Learnlet transform. Eventually, we investigate and compare the performance of different Unet architectures and Learnlet for image deconvolution in the astrophysical domain by following a two-step approach: a Tikhonov deconvolution with a closed-form solution, followed by post-processing with a neural network. To generate our training dataset, we extract HST cutouts from the CANDELS survey in the F606W filter (V-band) and corrupt these images to simulate their blurred-noisy versions. Our numerical results based on these simulations show a detailed comparison between the considered methods for different noise levels.

astro-ph.IM↗