Munich Health Foundation Model Symposium

Session

Break + Posters session

10 Apr 2024, 15:00

Auditorium, Building 23 (Helmholtz Munich Campus)

11. A generative recommender system with GMM prior for cancer drug generation and sensitivity prediction

Poster

Talk

Recent emergence of high-throughput drug screening assays sparkled an intensive development of machine learning methods, including models for prediction of sensitivity of cancer cell lines to anticancer drugs, as well as methods for generation of potential drug candidates. However, the concept of generating compounds with specific properties and simultaneous modeling of their efficacy against...

12. A generative recommender system with GMM prior for cancer drug generation and sensitivity prediction

Poster

Poster

Recent emergence of high-throughput drug screening assays sparkled an intensive development of machine learning methods, including models for prediction of sensitivity of cancer cell lines to anticancer drugs, as well as methods for generation of potential drug candidates. However, the concept of generating compounds with specific properties and simultaneous modeling of their efficacy against...

25. A Machine learning based model for Multitrait Allelic Series Scoring

Poster

Poster

Allelic series describe how the range of mutations within a single gene impacts gene function, and their study can help identify new candidate targets[1]. Despite the advances in machine learning strategies to model the expected impact of variants aggregating various sources of variant annotation [2][3], methods describing effects across multiple phenotypes are missing. To fill this gap, we...

47. Advancing Clinical Trials Through Digital Twins

Poster

Clinical trials play a fundamental role in medical research, serving as the bridge between the development of safe and effective medical interventions and their widespread use for public benefit. Despite their significance, the process of conducting clinical research is paved with challenges. It is a time-consuming task that requires substantial financial investment and yet exhibits alarmingly...

27. Advancing Genetic Analysis of Aging through Multivariate Brain Imaging Embeddings

Poster

Poster

The genetic underpinnings of brain structural changes with aging hold the key to understanding degenerative and neurological diseases. Traditional deep learning methods have typically approached aging by attempting to predict chronological age from brain MRI and using the resulting predictions as a phenotype, an approach that considers aging as a univariate phenomenon and thereby fails to...

28. An Interpretable Graph Message Passing Neural Network for predicting disease causing genes in viral infections

Poster

Talk

Viral infections are complex multisystemic diseases characterized by genetic and molecular alterations in the host. Consequently, Identifying disease-related genes and their underlying molecular mechanisms poses significant challenges. To address these challenges, we have enhanced our methodology by employing Graph Message Passing Neural Networks (MPNNs) that leverage host proteins as input...

19. BayeSDX: A Bayesian Test for Spatially Differentially Expressed Genes Across Slides

Poster

Spatial transcriptomics datasets, comprising slides from various subjects and conditions, facilitate the exploration of gene expression changes in health and disease within their spatial context. Despite this potential, statistical methods for analyzing differential spatial expression across subject groups are missing. Addressing this need, we introduce BayeSDX, a Bayesian test for Spatial...

35. DataDream: Few-shot Guided Dataset Generation

Poster

Talk

While text-to-image diffusion models have been shown to achieve state-of-the-art results in image synthesis, they have yet to prove their effectiveness in downstream applications. Previous work has proposed to generate data for image classifier training given limited real data access. However, these methods struggle to generate in-distribution images or depict fine-grained features, thereby...

14. Digital Diagnostik Copilot: Rethinking Diagnostic Decision Support Systems

Poster

Poster

The current diagnostic decision support systems approach falls into two categories: protocol compliance and disease-specific diagnostic models. Protocol compliance is designed to convert state-of-the-art diagnostic medicine into a procedural checklist, ensuring the diagnosing physician remembers all critical steps. In contrast, diagnostic models are designed to provide the likelihood of a...

33. DinoBloom: A Foundation Model for Generalizable Cell Embeddings in Hematology

Poster

Poster

In hematology, computational models offer significant potential
to improve diagnostic accuracy, streamline workflows, and reduce
the tedious work of analyzing single cells in peripheral blood or bone
marrow smears. However, clinical adoption of computational models has
been hampered by the lack of generalization due to large batch effects,
small dataset sizes, and poor performance in...

9. Discovering highly potent antimicrobial peptides with deep generative model HydrAMP

Poster

Talk

Antimicrobial peptides emerge as compounds that can alleviate the global health hazard of antimicrobial resistance, prompting a need for novel computational approaches to peptide generation. Here, we propose HydrAMP, a conditional variational autoencoder that learns lower-dimensional, continuous representation of peptides and captures their antimicrobial properties. The model disentangles the...

29. Discovery of immunological cellular neighborhoods from protein markers in spatial tumor data

Poster

Talk

Introduction

Spatial imaging of single cells and their protein markers in multiple tumor tissues offers detailed insights into the mechanisms of tumor-microenvironment interactions. The identification and characterization of cellular neighborhoods are vital to elucidating these mechanisms. However, existing approaches for cellular neighborhood analyses resort to predefined cell types or...

31. DNA language models provide a map of nucleotide and cis-regulatory code interactions in the genome

Poster

Talk

DNA and RNA are made up of contiguous chains of nucleotides, yet interactions between nucleotides are not strictly linear. Regions inside these sequences interact to form structures, define splicing, and establish cis-regulatory codes, among other genomic functions. DNA language models (DNA LMs) have been shown to capture local motifs and differentiate between regions in the genome with...

32. Efficient Vision Transformer for End-to-End Whole Slide Image Representation Learning in Cancer Subtyping Classification

Poster

Poster

Histopathology plays a central role in clinical medicine for tissue-based diagnostics and in biomedical research as a basis for understanding cellular and molecular mechanisms of disease. The microscopic evaluation of morphological information is a key component of computational pathology. The end-to-end whole slide image representation learning imposes a great challenge to the field of...

36. EgoCVR: An Egocentric Benchmark for Fine-Grained Composed Video Retrieval

Poster

Poster

In Composed Video Retrieval, a video and a textual description which modifies the video content are provided as inputs to the model. The aim is to retrieve the relevant video with the modified content from a database of videos. In this challenging task, the first step is to acquire large-scale training datasets and collect high-quality benchmarks for evaluation. In this work, we introduce...

2. Examining the effectiveness of foundation models on human 3’UTR sequences

Poster

Talk

Foundation models like DNABERT and Nucleotide Transformer have recently gained a lot of popularity in the field of DNA research. Adopted from Natural Language Processing, these models are trained in a self-supervised manner on vast amounts of genomic data. Once trained, foundation models offer applications for various downstream tasks, including promoter and enhancer prediction, prediction of...

18. Factor analysis and correlated topic model for multi-modal data integration

Poster

Poster

Integrating information across various data modalities can be beneficial for gaining valuable insights into underlying phenomena. Numerous methods exist for multi-modal data integration, ranging from linear matrix factorization-based approaches to nonlinear methods employing i.e. deep generative models. However, integrating data becomes particularly challenging when one or more modalities...

37. Foundation model for human heart simulation and analysis of the pathological changes

Poster

Talk

Nowadays, cardiologists and even practitioners are using different methods to evaluate the state of human heart. However, these possibilities do not mean that the patient is always secure during the diagnosis process. One of the important procedures is Right Heart Catheterization (RHC) that is consumed during the diagnosis process of pulmonary hypertension (PH). However, the problem with this...

22. Functional gene embedding improves rare variant polygenic risk scores

Poster

Talk

A key challenge in precision medicine is delineating how genetic variations determine phenotype. Studies show that individuals with low polygenic risk scores (PRS) for complex traits, carry more rare variants in relevant genes. Current genotype-to-phenotype models, such as PRS, have limited power in discovering associations between traits and genes with rare variants. Although rare variant...

4. Generating highly accurate pathology reports from gigapixel whole slide images with HistoGPT

Poster

Talk

Histopathology is considered the gold standard for determining the presence and nature of disease, particularly cancer. However, the process of analyzing tissue samples and producing a final pathology report is time-consuming, labor-intensive, and non-standardized. Therefore, new technological solutions are being sought to reduce the workload of pathologists. In this work, we present HistoGPT,...

21. HistoGWAS: A Framework for Genetic Analyses of Histological Images

Poster

Talk

Genetic analysis of medical imaging traits can help uncover disease mechanisms. Although digital pathology datasets offer vast opportunities for these analyses, methods for genetic analysis of such datasets are lacking. To fill this gap, we introduce HistoGWAS, a toolkit for genome-wide association studies (GWAS) of histological images. HistoGWAS utilizes semantic autoencoders to generate...

8. Insights from Acquiring Open Medical Imaging Datasets for Foundation Model Development

Poster

Talk

Foundation Models require significantly more data for training than earlier AI generations. The scarcity of clinical data, as well as the necessity of perfecting model generalization capabilities, make it necessary to aggregate data for model training and validation from various datasets. In this work, we explore the challenges related to FAIR clinical imaging data (findability, accessibility,...

7. Joint cell type identification in spatial transcriptomics and single-cell RNA sequencing data

Poster

Talk

Understanding the intricate composition of tissues in complex living organisms is crucial for unraveling the mechanisms underlying health and disease. This study addresses the challenge of dissecting cell types within tissues by integrating information from two powerful experimental techniques: single-cell RNA-sequencing (scRNA-seq) and spatial transcriptomics (ST). While scRNA-seq offers...

6. Joint Molecule Generation and Property Prediction with Jointformer

Poster

Poster

Modeling the joint distribution of the data samples and their properties promises to excel both in new data generation and property prediction, with capabilities reaching beyond separate generative or predictive models. However, training such joint models presents daunting challenges. Here, we propose to combine a transformer decoder with a transformer encoder in a single model that at the...

3. Large-scale characterization of cell niches in spatial atlases using bio-inspired graph learning

Poster

Poster

Spatial omics holds great potential to elucidate tissue architecture by dissecting underlying cell niches and cellular interactions. However, we lack an end-to-end computational framework that can effectively integrate different spatial omics tissue samples, quantitatively characterize cell niches based on biological knowledge of cell-cell communication and transcriptional regulation pathways,...

15. Leveraging Large Language Models to Enrich Clone Selection and Characterization in Antibody Discovery

Poster

Poster

The quest for efficient clone selection in B-cell research is pivotal to therapeutic antibody development. This study harnesses the power of a BERT-based large language model (LLM) trained on pair-chain antibodies, to analyse immune repertoires and facilitate clonal selection in addition to the established B-cell V-region recovery workflow (BCW). Two datasets were investigated in this study:...

5. Medical Reasoning in Large Language Models: A Chain of Thought analysis

Poster

Poster

With the growing popularity of AI chatbots, many people are querying large language models (LLMs) for medical advice. However, the decision-making of these models is hardly understood, and model hallucinations are particularly dangerous in the medical context. This work systematically analyzes medical reasoning in LLMs. We collect decision graphs from the medical literature and construct chain...

30. Metric Space Magnitude for Evaluating the Diversity of Latent Representations

Poster

Poster

The magnitude of a metric space is a recently-established invariant, providing a measure of the 'effective size' of a space across multiple scales while also capturing numerous geometrical properties. We develop a family of magnitude-based measures of the intrinsic diversity of latent representations. Our measures are provably stable under perturbations of the data, can be efficiently...

26. Mixed Models with Multiple Instance Learning

Poster

Talk

Predicting patient features from single-cell data can help identify cellular states implicated in health and disease. Linear models and average cell type expressions are typically favored for this task for their efficiency and robustness, but they overlook the rich cell heterogeneity inherent in single-cell data. To address this gap, we introduce MixMIL, a framework integrating Generalized...

10. ML-enabled genetic causal inference of biological aging

Poster

Poster

The aging process is a complex phenomenon intricately linked with the onset and progression of various age-related diseases. In this study, we propose a novel framework that synergistically combines machine learning techniques with genetic causal inference methods to elucidate the underlying mechanisms of biological aging and its association with disease susceptibility.
Our research employs...

20. Pre-trained molecular representations enable antimicrobial discovery

Poster

Poster

The rise in antimicrobial resistance poses a worldwide threat, reducing the efficacy of common antibiotics. Yet, determining the antimicrobial activity of new chemical compounds through experimental methods is still a time-consuming and costly endeavor.
Compound-centric deep learning models hold the promise to speed up the
search and prioritization process. Here, we introduce a lightweight...

49. Probabilistic modeling of tumor infiltration processes

Poster

Talk

Introduction: Spatial molecular data play a crucial role in understanding the tumor immune microenvironment (TIME). Yet, there is a notable absence of efficient computational techniques for tracking infiltration patterns within cancer tissue. Our work aims to address this gap. Through a graph-based and probabilistic modeling-based approach to spatial data representation, we describe short-...

13. Probabilistic quantification of high-dimensional structural features from tomographic sensing data

Poster

Poster

Biosensing technologies capture information on tissue in a time- and cost-efficient way. However, many complex, high-dimensional biomarkers that can be obtained from tomographic imaging cannot be easily extracted from sensing data due to the low dimensionality and the limited amount of information recorded.
We introduce an AI-empowered framework to quantify multi-dimensional tissue...

38. Towards foundation models of human cognition

Poster

Talk

We are building a foundation model of human cognition building on our earlier work [1]. For this, we have already collected a large-scale data set by transcribing nearly 200 psychological experiments into a text-based form. This data set reaches an unprecedented scale, consisting of nearly 20 million human choices from about 300,000 participants, covering domains such as decision-making,...

48. Unbalancedness in Neural Monge Maps Improves Unpaired Domain Translation

Poster

In optimal transport (OT), a Monge map is known as a mapping that transports a source distribution to a target distribution in the most cost-efficient way. Recently, multiple neural estimators for Monge maps have been developed and applied in diverse unpaired domain translation tasks, e.g. in single-cell biology and computer vision. However, the classic OT framework enforces mass conservation,...

24. Uncovering early single-cell marker of Eczema with scMR

Poster

Poster

Eczema is the most common inflammatory skin disease, with a prevalence of 25% in school-aged children and 10% in adults (Brown, 2016). Despite its high incidence, molecular drivers are still unknown. Revealing these causal risk factors provides essential disease onset prediction and prevention opportunities, e.g., targeting them with new and repurposed drugs.

Mendelian Randomization (MR)...

23. Understanding the epithelial-to-mesenchymal transition in cancer with a single-cell large language model

Poster

Poster

The epithelial-to-mesenchymal transition (EMT) is pivotal in tumour progression and resistance to treatment, yet its heterogeneity complicates the precise assessment of EMT status of individual tumour cells in different cancer types. Furthermore, while key epithelial and mesenchymal genes driving the transformation are well characterised, other regulators, especially at intermediate stages of...

16. Unraveling the foundation of T cell receptors

Poster

Poster

Deciphering the interaction between T-cell receptors (TCRs) and disease-derived epitopes has been declared as one of nine Cancer Grand Challenges in 2023. This interaction plays a pivotal role in the immune system's ability to recognize and eliminate infectious pathogens and cancerous cells. Sequence-based machine learning models for predicting TCR-epitope binding are constrained by limited...

34. VLMGender: A Benchmark for Evaluating Gender Bias in Vision-Language Assistants

Poster

Talk

Pre-trained large language models (LLMs) have been reliably integrated with visual input for multimodal tasks. The widespread adoption of instruction-tuned image-to-text vision-language assistants (VLAs) like LLaVA and MiniGPT, necessitates evaluating social biases. We propose a comprehensive prompt-image benchmark measuring gender bias for VLAs in personality traits, skills, and occupations....

17. Whole Genome Transformers for Gene Interaction Effects in Microbiome Habitat Prediction

Poster

Poster

Leveraging the vast genetic diversity within microbiomes offers unparalleled insights into complex phenotypes, yet the task of accurately predicting and understanding such traits from genomic data remains challenging. We propose a framework that leverages existing large models for gene vectorization to predict habitat specificity from entire microbial genome sequences. Based on our model, we...