Searcharxiv⌕ Search

arXiv subjects

Sanne Abeln

Publications and source records attributed to Sanne Abeln.

At least 19 recordsLinked to original sources
arXiv2026

Bridging the gap between Performance and Interpretability: An Explainable Disentangled Multimodal Framework for Cancer Survival Prediction

While multimodal survival prediction models are increasingly more accurate, their complexity often reduces interpretability, limiting insight into how different data sources influence predictions. To address this, we introduce DIMAFx, an explainable multimodal framework for cancer survival prediction that produces disentangled, interpretable modality-specific and modality-shared representations from histopathology whole-slide images and transcriptomics data. Across multiple cancer cohorts, DIMAFx achieves state-of-the-art performance and improved representation disentanglement. Leveraging its interpretable design and SHapley Additive exPlanations, DIMAFx systematically reveals key multimodal interactions and the biological information encoded in the disentangled representations. In breast cancer survival prediction, the most predictive features contain modality-shared information, including one capturing solid tumor morphology contextualized primarily by late estrogen response, where higher-grade morphology aligned with pathway upregulation and increased risk, consistent with known breast cancer biology. Key modality-specific features capture microenvironmental signals from interacting adipose and stromal morphologies. These results show that multimodal models can overcome the traditional trade-off between performance and explainability, supporting their application in precision medicine.

cs.CV↗
arXiv2025

Exploring Molecular Odor Taxonomies for Structure-based Odor Predictions using Machine Learning

One of the key challenges to predict odor from molecular structure is unarguably our limited understanding of the odor space and the complexity of the underlying structure-odor relationships. Here, we show that the predictive performance of machine learning models for structure-based odor predictions can be improved using both, an expert and a data-driven odor taxonomy. The expert taxonomy is based on semantic and perceptual similarities, while the data-driven taxonomy is based on clustering co-occurrence patterns of odor descriptors directly from the prepared dataset. Both taxonomies improve the predictions of different machine learning models and outperform random groupings of descriptors that do not reflect existing relations between odor descriptors. We assess the quality of both taxonomies through their predictive performance across different odor classes and perform an in-depth error analysis highlighting the complexity of odor-structure relationships and identifying potential inconsistencies within the taxonomies by showcasing pear odorants used in perfumery. The data-driven taxonomy allows us to critically evaluate our expert taxonomy and better understand the molecular odor space. Both taxonomies as well as a full dataset are made available to the community, providing a stepping stone for a future community-driven exploration of the molecular basis of smell. In addition, we provide a detailed multi-layer expert taxonomy including a total of 777 different descriptors from the Pyrfume repository.

q-bio.QM↗
arXiv2025

Explainable AI in Healthcare: to Explain, to Predict, or to Describe?

Explainable Artificial Intelligence (AI) methods are designed to provide information about how AI-based models make predictions. In healthcare, there is a widespread expectation that these methods will provide relevant and accurate information about a model's inner-workings to different stakeholders (ranging from patients and healthcare providers to AI and medical guideline developers). This is a challenging endeavour since what qualifies as relevant information may differ greatly depending on the stakeholder. For many stakeholders, relevant explanations are causal in nature, yet, explainable AI methods are often not able to deliver this information. Using the Describe-Predict-Explain framework, we argue that Explainable AI methods are good descriptive tools, as they may help to describe how a model works but are limited in their ability to explain why a model works in terms of true underlying biological mechanisms and cause-and-effect relations. This limits the suitability of explainable AI methods to provide actionable advice to patients or to judge the face validity of AI-based models.

stat.ME↗
arXiv2025

Beyond Olfaction: New Insights into Human Odorant Binding Proteins

Until today, the exact function of mammalian odorant binding proteins (OBPs) remains a topic of debate. Although their main established function lacks direct evidence in human olfaction, OBPs are traditionally believed to act as odorant transporters in the olfactory sense, which led to the exploration of OBPs as biomimetic sensor units in artificial noses. Now, available RNA-seq and proteomics data identified the expression of human OBPs (hOBP2A and hOBP2B) in both, male and female reproductive tissues. This observation prompted the conjecture that OBPs may possess functions that go beyond the olfactory sense, potentially as hormone transporters. Such a function could further link them to the tumorigenesis and cancer progression of hormone dependent cancer types including ovarian, breast, prostate and uterine cancer. In this structured review, we use available data to explore the effects of genetic alterations such as somatic copy number aberrations and single nucleotide variants on OBP function and their corresponding gene expression profiles. Our computational analyses suggest that somatic copy number aberrations in OBPs are associated with large changes in gene expression in reproductive cancers while point mutations have little to no effect. Additionally, the structural characteristics of OBPs, together with other lipocalin family members, allow us to explore putative functions within the context of cancer biology. Our overview consolidates current knowledge on putative human OBP functions, their expression patterns, and structural features. Finally, it provides an overview on applications, highlighting emerging hypotheses and future research directions within olfactory and non-olfactory roles.

q-bio.BM↗
arXiv2025

Disentangled and Interpretable Multimodal Attention Fusion for Cancer Survival Prediction

To improve the prediction of cancer survival using whole-slide images and transcriptomics data, it is crucial to capture both modality-shared and modality-specific information. However, multimodal frameworks often entangle these representations, limiting interpretability and potentially suppressing discriminative features. To address this, we propose Disentangled and Interpretable Multimodal Attention Fusion (DIMAF), a multimodal framework that separates the intra- and inter-modal interactions within an attention-based fusion mechanism to learn distinct modality-specific and modality-shared representations. We introduce a loss based on Distance Correlation to promote disentanglement between these representations and integrate Shapley additive explanations to assess their relative contributions to survival prediction. We evaluate DIMAF on four public cancer survival datasets, achieving a relative average improvement of 1.85% in performance and 23.7% in disentanglement compared to current state-of-the-art multimodal models. Beyond improved performance, our interpretable framework enables a deeper exploration of the underlying interactions between and within modalities in cancer biology.

cs.CV↗
arXiv2025

PLM-eXplain: Divide and Conquer the Protein Embedding Space

Protein language models (PLMs) have revolutionised computational biology through their ability to generate powerful sequence representations for diverse prediction tasks. However, their black-box nature limits biological interpretation and translation to actionable insights. We present an explainable adapter layer - PLM-eXplain (PLM-X), that bridges this gap by factoring PLM embeddings into two components: an interpretable subspace based on established biochemical features, and a residual subspace that preserves the model's predictive power. Using embeddings from ESM2, our adapter incorporates well-established properties, including secondary structure and hydropathy while maintaining high performance. We demonstrate the effectiveness of our approach across three protein-level classification tasks: prediction of extracellular vesicle association, identification of transmembrane helices, and prediction of aggregation propensity. PLM-X enables biological interpretation of model decisions without sacrificing accuracy, offering a generalisable solution for enhancing PLM interpretability across various downstream applications. This work addresses a critical need in computational biology by providing a bridge between powerful deep learning models and actionable biological insights.

q-bio.BM↗
arXiv2024

Mimicking the Gas-Phase to Transport Odorants through the Nasal Mucus: Functional Insights into Odorant Binding Proteins

Mammalian odorant binding proteins (OBPs) have long been suggested to transport hydrophobic odorant molecules through the aqueous environment of the nasal mucus. While the function of OBPs as odorant transporters is supported by their hydrophobic beta-barrel structure, no rationale has been provided on why and how these proteins facilitate the uptake of odorants from the gas phase. Here, a multi-scale computational approach validated through available high-resolution spectroscopy experiments reveals that the conformational space explored by carvone inside the binding cavity of porcine OBP (pOBP) is much closer to the gas than the aqueous phase, and that pOBP effectively manages to transport odorants by lowering the free energy barrier of odorant uptake. Understanding such perireceptor events is crucial to fully unravel the molecular processes underlying the olfactory sense, and move towards the development of protein-based biomimetic sensor units that can serve as artificial noses.

q-bio.BM↗
arXiv2024

PatchProt: Hydrophobic patch prediction using protein foundation models

Hydrophobic patches on protein surfaces play important functional roles in protein-protein and protein-ligand interactions. Large hydrophobic surfaces are also involved in the progression of aggregation diseases. Predicting exposed hydrophobic patches from a protein sequence has been shown to be a difficult task. Fine-tuning foundation models allows for adapting a model to the specific nuances of a new task using a much smaller dataset. Additionally, multi-task deep learning offers a promising solution for addressing data gaps, simultaneously outperforming single-task methods. In this study, we harnessed a recently released leading large language model ESM-2. Efficient fine-tuning of ESM-2 was achieved by leveraging a recently developed parameter-efficient fine-tuning method. This approach enabled comprehensive training of model layers without excessive parameters and without the need to include a computationally expensive multiple sequence analysis. We explored several related tasks, at local (residue) and global (protein) levels, to improve the representation of the model. As a result, our fine-tuned ESM-2 model, PatchProt, cannot only predict hydrophobic patch areas but also outperforms existing methods at predicting primary tasks, including secondary structure and surface accessibility predictions. Importantly, our analysis shows that including related local tasks can improve predictions on more difficult global tasks. This research sets a new standard for sequence-based protein property prediction and highlights the remarkable potential of fine-tuning foundation models enriching the model representation by training over related tasks.

q-bio.QM↗
arXiv2024

CIBRA identifies genomic alterations with a system-wide impact on tumor biology

Background: Genomic instability is a hallmark of cancer, leading to many somatic alterations. Identifying which alterations have a system-wide impact is a challenging task. Nevertheless, this is an essential first step for prioritizing potential biomarkers. We developed CIBRA (Computational Identification of Biologically Relevant Alterations), a method that determines the system-wide impact of genomic alterations on tumor biology by integrating two distinct omics data types: one indicating genomic alterations (e.g., genomics), and another defining a system-wide expression response (e.g., transcriptomics). CIBRA was evaluated with genome-wide screens in 33 cancer types using primary and metastatic cancer data from the Cancer Genome Atlas and Hartwig Medical Foundation. Results: We demonstrate the capability of CIBRA by successfully confirming the impact of point mutations in experimentally validated oncogenes and tumor suppressor genes. Surprisingly, many genes affected by structural variants were identified to have a strong system-wide impact (30.3%), suggesting that their role in cancer development has thus far been largely underreported. Additionally, CIBRA can identify impact with only ten cases and controls, providing a novel way to prioritize genomic alterations with a prominent role in cancer biology. Conclusions: Our findings demonstrate that CIBRA can identify cancer drivers by combining genomics and transcriptomics data. Moreover, our work shows an unexpected substantial system-wide impact of structural variants in cancer. Hence, CIBRA has the potential to preselect and refine current definitions of genomic alterations to derive more nuanced biomarkers for diagnostics, disease progression, and treatment response. CIBRA is available at https://github.com/AIT4LIFE-UU/CIBRA

q-bio.GN↗
arXiv2023

Structural Property Prediction

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics. Some structural properties of proteins that are closely linked to their function may be easier (or much faster) to predict from sequence than the complete tertiary structure; for example, secondary structure, surface accessibility, flexibility, disorder, interface regions or hydrophobic patches. Serving as building blocks for the native protein fold, these structural properties also contain important structural and functional information not apparent from the amino acid sequence. Here, we will first give an introduction into the application of machine learning for structural property prediction, and explain the concepts of cross-validation and benchmarking. Next, we will review various methods that incorporate knowledge of these concepts to predict those structural properties, such as secondary structure, surface accessibility, disorder and flexibility, and aggregation.

q-bio.BM↗
arXiv2023

Preface to Introduction to Protein Structural Bioinformatics

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics.

q-bio.BM↗
arXiv2023

Introduction to Protein Structure

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics. Within the living cell, protein molecules perform specific functions, typically by interacting with other proteins, DNA, RNA or small molecules. They take on a specific three dimensional structure, encoded by its amino acid sequence, which allows them to function within the cell. Hence, the understanding of a protein's function is tightly coupled to its sequence and its three dimensional structure. Before going into protein structure analysis and prediction, and protein folding and dynamics, here, we give a short and concise introduction into the basics of protein structures.

q-bio.BM↗
arXiv2023

Structure Alignment

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics. The Protein DataBank (PDB) contains a wealth of structural information. In order to investigate the similarity between different proteins in this database, one can compare the primary sequence through pairwise alignment and calculate the sequence identity (or similarity) over the two sequences. This strategy will work particularly well if the proteins you want to compare are close homologs. However, in this chapter we will explain that a structural comparison through structural alignment will give you much more valuable information, that allows you to investigate similarities between proteins that cannot be discovered by comparing the sequences alone.

q-bio.BM↗
arXiv2023

Data Resources for Structural Bioinformatics

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics. Structural bioinformatics involves a variety of computational methods, all of which require input data. Typical inputs include protein structures and sequences, which are usually retrieved from a public or private database. This chapter introduces several key resources that make such data available, as well as a handful of tools that derive additional information from experimentally determined or computationally predicted protein structures and sequences.

q-bio.BM↗
arXiv2023

Function Prediction

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics. There are still huge gaps in understanding the molecular function of proteins. This raises the question on how we may predict protein function, when little to no knowledge from direct experiments is available. Protein function is a broad concept which spans different scales: from quantum scale effects for catalyzing enzymatic reactions, to phenotypes that manifest at the organism level. In fact, many of these functional scales are entirely different research areas. Here, we will consider prediction of a smaller range of functions, roughly spanning the protein residue-level up to the pathway level. We will give a conceptual overview of which functional aspects of proteins we can predict, which methods are currently available, and how well they work in practice.

q-bio.BM↗
arXiv2023

Introduction to Protein Folding

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics. In this chapter we explore basic physical and chemical concepts required to understand protein folding. We introduce major (de)stabilising factors of folded protein structures such as the hydrophobic effect and backbone entropy. In addition, we consider different states along the folding pathway, as well as natively disordered proteins and aggregated protein states. In this chapter, an intuitive understanding is provided about the protein folding process, to prepare for the next chapter on the thermodynamics of protein folding. In particular, it is emphasized that protein folding is a stochastic process and that proteins unfold and refold in a dynamic equilibrium. The effect of temperature on the stability of the folded and unfolded states is also explained.

q-bio.BM↗
arXiv2023

Thermodynamics of Protein Folding

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics. In the previous chapter, "Introduction to Protein Folding", we introduced the concept of free energy and the protein folding landscape. Here, we provide a deeper, more formal underpinning of free energy in terms of the entropy and enthalpy; to this end, we will first need to better define the meaning of equilibrium, entropy and enthalpy. When we understand these concepts, we will come back for a more quantitative explanation of protein folding and dynamics. We will discuss the influence of temperature on the free energy landscape, and the difference between microstates and macrostates.

q-bio.BM↗
arXiv2023

Molecular Dynamics

While many good textbooks are available on Protein Structure, Molecular Simulations, Thermodynamics and Bioinformatics methods in general, there is no good introductory level book for the field of Structural Bioinformatics. This book aims to give an introduction into Structural Bioinformatics, which is where the previous topics meet to explore three dimensional protein structures through computational analysis. We provide an overview of existing computational techniques, to validate, simulate, predict and analyse protein structures. More importantly, it will aim to provide practical knowledge about how and when to use such techniques. We will consider proteins from three major vantage points: Protein structure quantification, Protein structure prediction, and Protein simulation & dynamics. We know that many proteins have functional motions, and in Chapter "Structure Determination" we already introduced the famous example of the allosteric cooperative binding of oxygen to the haem group in hemoglobin. However, experimentally, such motions are hard to observe. Here, we will introduce MD simulations to investigate the dynamic behaviour of proteins. In a simulation the forces and interactions between particles are used to numerically derive the resulting three-dimensional movement of these particles over a certain time-scale. We will also highlight some applications, and will see how simulation results may be interpreted.

q-bio.BM↗