Accurate prediction of the temporal dynamics of biological systems is crucial for informing timely and effective interventions, e.g., in ecological or epidemiological contexts, or for treatment adjustments in therapy. While machine learning has proven its capabilities in generalizing the underlying non-linear dynamics of such systems, unlocking its predictive power is often restrained by the limited availability of large, curated datasets. To supplement real-world data, informing machine learning by transfer learning with synthetic data derived from simulations using ordinary differential equations has emerged as a promising solution. However, the success of this approach highly depends on the designed characteristics of the synthetic data. We suggest scrutinizing these characteristics, such as size, diversity, and noise, of ordinary differential equation-based synthetic time series datasets. Here, we demonstrate how to systematically evaluate the influence of such design choices on transfer learning performance. We conduct a proof-of-concept study on three simple, but widely used systems and four real-world datasets. We find a strong interdependency between synthetic dataset size and diversity effects. Good transfer learning settings heavily rely on real-world data characteristics as well as the data's coherence with the dynamics of the model underlying the synthetic data. We achieve a performance improvement of up to 95% in mean absolute error for simulation-based transfer learning compared to non-informed deep learning. Our work emphasizes the relevance of carefully selecting properties of synthetic data for leveraging the valuable domain knowledge contained in ordinary differential equation models for machine-learning based predictions. The code is available at https://github.com/DILiS-lab/opt-synthdata-4tl .

Topological domains(TADs) are consecutive genomic locations with denser local interactions to a certain extent, and they are important for cellular gene expression control and modulation. TADs were first identified when studying three-dimensional genomic structures over Hi-C interaction datasets. Many studies have focused on developing approaches in inferring TADs, which led to multiple TAD-caller approaches development. On the other hand, the number of RNA interactome datasets, such as RNA-RNA and RNA-DNA interactions, has recently been increasing. Even though TADs have been extensively studied in Hi-C datasets, they have not been studied across these RNA interactomes. We conducted a systematized comparison of 28 TAD-callers across mammalians over RNA-associated interactomes(RAIs), especially RNA-RNA and RNA-DNA interaction datasets at a high resolution. Our findings highlighted the significant enrichment of Cohesin/CTCF proteins at RNA-TAD boundaries for both RNA-RNA and RNA-DNA interactomes, especially those with corner dots, which are similar to the results for original TADs. The sizes and numbers of RNA-TADs vary significantly between different TAD caller approaches and RNA interactome resolution, suggesting the importance of considering RNA-TADs as hierarchical domains rather than distinct intervals. We examined the core principles and assumptions behind TAD-callers over RNA interactomes. To our best knowledge, this is the first time that many TAD inference methods are adapted to infer TAD-like domains on RNAs. Our results provide valuable guidance in selecting the most suitable methods for TAD inference over RNA interactomes.

Accurate prediction of MHC-peptide binding affinity remains a challenge for immunotherapeutic development. Existing methods struggle to jointly model functional semantics of polymorphic residues, evolutionary conservation constraints, and structural dynamic. We propose the Contrast learning-based Multi-feature Heterogeneous Subgraph model (CMHS) with sequence and structural representation. For sequence representation, we introduce LoRA fine-tuning to obtain the MHC-exclusive sequence representation from ESM2, then jointly BLOSUM50 to capture long-range functional dependencies and evolutionarily conserved residues. For structural representation, we use the biophysics-guided heterogeneous graph network. Constructing an MHC-peptide graph with a novel trainable Gaussian noise layer guided by crystallographic B-factors to dynamically simulate electron density uncertainty, coupled with a three-stage message-passing framework with subgraph aggregation, subgraph extraction and heterogeneous. Finally, to align sequence and graph representation spaces, we use contrastive learning to obtain a more comprehensive representation and to enhance the ability of model prediction. Evaluations on 16 HLA allele benchmarks show average SRCC improvements of 8.7%, with improvements of average AUC of 7.6%. This work establishes a new paradigm for predicting hypervariable immune interactions. The corresponding code can be founded in github.

Gut microbiome dysbiosis is a critical determinant for autologous fecal microbiota transplantation (Auto-FMT) eligibility, yet current classification approaches rely predominantly on supervised learning with manually annotated sequencing labels, which are often scarce. This study proposes DynaBiome, a framework designed to predict gut dysbiosis by leveraging unsupervised learning and clinical phenotypic proxies as a scalable alternative to ground-truth genomic labeling. Our framework employs an LSTM autoencoder architecture to capture temporal microbiome dynamics within 14-day windows. The model reconstructs normal microbiome patterns, where high reconstruction errors signal potential dysbiosis. To ensure rigorous evaluation and prevent data leakage, the dataset was partitioned via a strict patient-level split. Unsupervised anomaly signals were refined via phenotypic proxy labels (e.g., fever, neutropenia) via weak supervision, and ensemble learning methods were applied to optimize classification performance. The initial LSTM autoencoder successfully flagged dysbiotic sequences but required refinement to reduce false positives. Ensemble learning significantly enhanced predictive accuracy. The stacked ensemble (with Logistic Regression meta-learner) demonstrated optimal performance with an ROC AUC of 0.8908 and a Weighted F1-score of 0.7909. This approach significantly outperformed the standard One-Class SVM baseline (ROC AUC 0.6033), confirming the superiority of deep temporal modeling over static anomaly detection. Critically, the model achieved performance levels comparable to fully supervised baselines, confirming the efficacy of the proxy-label framework. Integrating unsupervised temporal feature extraction with stacked ensemble methods provides a viable framework for dysbiosis prediction. These results demonstrate that leveraging phenotypic via weak supervision can effectively approximate supervised baselines, thereby reducing the reliance on comprehensive metagenomic annotations for longitudinal patient monitoring.

Integrating genomic datasets from homogenous or disparate sources to identify genes that are commonly or uniquely expressed remains a largely underexplored area. Such integrative analysis can reveal biologically relevant genes that are common or exclusive across datasets or within specific conditions or cohorts. Identifying these gene expression profiles and employing them to classify disease status can aid in the development of vaccines, diagnostics and targeted therapeutics with efficacy against difficult-to-treat medically important pathogens and cancer. This work develops new methodologies to integrate transcriptomic patterns from the lungs and spleen tissues infected by Francisella tularensis – Schu4 and Live Vaccine Strain (LVS). Our objective is to (i) identify biologically relevant gene features indicative of respiratory infection, disease severity, and bacterial dissemination to the spleen, and (ii) develop a Weighted $$\ell _1$$-norm Non-Parallel Support Vector Machines ($$\ell _1$$-WNPSVM) that will utilize the selected genes to predict disease status. The $$\ell _1$$-WNPSVM is trained on the lungs data and validated on the spleen data, introducing a form of transfer learning, with uninfected controls and Schu4 or LVS samples as classes. Currently, a direct application of existing NPSVM-type methods to analyze gene expression datasets, where the number of genes significantly exceeds the number of samples, is computationally impractical due to their large memory requirements. This work addresses these challenges and also generalizes to models of similar formulations by incorporating dimensionality reduction and gene selection into the NPSVM-type frameworks. The $$\ell _1$$-WNPSVM method outperforms traditional machine learning techniques such as ANN, XGBoost, AdaBoost, GradBoost, KNN, SVM, Naive Bayes, Random Forest, Logistic Regression, and Decision Tree, achieving a $$97\%$$ balanced accuracy on imbalanced data. We discovered sets of 235 genes exclusively expressed in the lungs and spleen tissues and utilized them to classify bacterial strains and controls, enabling prediction of disease status. Gene ontology is performed to reveal underlying metabolic pathways. Our analysis shows that signal transduction and disease (cancer) pathways are the most significant pathways activated in the lungs while gene expression (transcription), immune system, and disease (cancer) pathways are activated in the spleen. Collectively, these pathways indicate a significant host response to infection, including how the bacteria interact with host tissues during dissemination.

Cryo-electron microscopy (cryo-EM) has emerged as a powerful technique for high-resolution structural determination of macromolecules. However, accurately classifying single-particle cryo-EM images remains challenging, especially when dealing with deformed particles. In traditional 2D classification methods, clustering algorithms are used for classification. This assumption leads to some deformed particles being misclassified in 2D images, which adversely affects downstream tasks. To address this challenge, we propose a point cloud-based deformation measurement model that integrates a Variational Autoencoder (VAE) with a heuristic point cloud matching algorithm to calculate particle deformation values. This model enables the identification and removal of particles with large deformations. Our experiments on simulated and real cryo-EM datasets, including Tobacco Mosaic Virus (TMV) and mixed capsids of MS2 virions (MS2). The model achieves robust classification (F1: 0.85-0.88) while preserving 93-95% of structural details, and can effectively filter out deformed particles after 2D classification. The model identifies and removes deformed or misclassified particles to improve classification quality. It serves as a data-filtering post-processing step following 2D classification. By improving the quality of particle datasets, it enhances the reliability of subsequent analysis in cryo-EM.

Existing tools for virus variant identification can pinpoint the most abundant virus variant in a sequencing sample. However, patients can be infected by more than one variant of the same virus species or strain, for example by multiple variants of SARS-CoV-2. This leads to the more complicated problem of virus variant quantification from samples containing virus mixtures. We report on improvements of Orthanq, our generic tool for haplotype quantification, and show how it can be applied to perform uncertainty aware quantification of virus variants. We evaluate this ability on simulated and real SARS-CoV-2 and HIV-1 virus mixture datasets and show that Orthanq outperforms other state of the art approaches. Orthanq performs identification and uncertainty-aware quantification of known virus variants of any virus species, in particular also in samples with mixed infections. With extensive built-in visualizations and reporting of alternative solutions with posterior densities, users can easily evaluate the uncertainty of the results.

DNA sequences are fundamental carriers of genetic information, and their accurate classification is essential for understanding gene regulation, disease mechanisms, and translational genomics. Existing encoding methods often fail to capture both local and long-range dependencies simultaneously. We introduce EDEN (Expected Density of Nucleotide Encoding), a unified multiscale encoding framework based on kernel density estimation. EDEN captures position-specific and context-dependent nucleotide patterns and integrates them into a hybrid deep learning architecture. Across sixteen benchmark datasets covering promoter detection, core promoter detection, and transcription factor binding prediction, EDEN achieves the best average performance while using orders of magnitude fewer parameters compared with state-of-the-art models. All source code, pretrained models, and datasets are publicly available at: https://github.com/zabihis/EDEN . EDEN provides an efficient, biologically informed, and interpretable multiscale representation for genomic sequence classification. Its favorable parameter-performance ratio and robust consistency across tasks underscore its practicality for large-scale genomic applications.