Günter Klambauer

We present VN-EGNN, a novel approach to binding site identification that significantly advances predictive performance. By integrating virtual nodes into E

Base editing shows great potential in research and clinical applications. Current iterations of the deaminases used to create precise single-nucleotide changes via base editing exhibit undesirable effects, including off-targeting, off-base editing, and bystander editing. Current deaminases are derived from either larger eukaryotic deaminases, which exhibit high levels of Cas-independent DNA targeting, or from evolved variants of the smaller E. coli TadA protein (ecTadA), which exhibits off-base editing. To overcome the limitations inherent to using a single protein sequence for engineering, we diversified newly identified TadA orthologs by DNA shuffling to yield millions of training sequences for measuring base editor efficiency. We trained generative models on the performance data from the pools of variants and drew on information-theoretic insights to efficiently explore the sequence space to generate diverse and high-performing deaminases. From a single round of diversification, we created a small set of novel and specific cytosine and adenosine deaminases that were markedly distinct in sequence from published base editor deaminases. We found that our model created deaminases generally outperform those we identified through typical directed evolution. The novel compact deaminases identified here show high on-base activity, comparable to the leading published base editors, and with demonstrably lower off-base activity.

Today’s drug discovery increasingly relies on computational and machine learning approaches to identify novel candidates, yet data scarcity remains a significant challenge. To address this limitation, we present MHNfs, an application specifically designed to predict molecular activity in low-data scenarios. At its core, MHNfs leverages a state-of-the-art few-shot activity prediction model, named MHNfs, which has demonstrated strong performance across a large set of prediction tasks in the benchmark data set FS-Mol. The application features an intuitive interface that enables users to prompt the model for precise activity predictions based on a small number of known active and inactive molecules, akin to interactive interfaces for large language models. To evaluate its efficacy, we simulate real-world scenarios by recasting PubChem bioassays as few-shot prediction tasks. MHNfs offers a streamlined and accessible solution for deploying advanced few-shot learning models, providing a valuable tool for accelerating drug discovery.

Type 1 Diabetes (T1D) is a T-cell mediated disease with a strong immunogenetic HLA dependence. HLA allelic influence on the T cell receptor (TCR) repertoire shapes thymic selection and controls activation of diabetogenic clones, yet remains largely unresolved in T1D. We sequenced the circulating TCRβ chain repertoire from 2250 HLA-typed individuals across three cross-sectional cohorts, including T1D patients, and healthy related and unrelated controls. We found that HLA risk alleles show higher restriction of TCR repertoires in T1D individuals. We leveraged deep learning to identify T1D-associated TCR subsequence motifs that were also observed in independent TCR cohorts residing in pancreas-draining lymph nodes of T1D individuals. Collectively, our data demonstrate T1D-related TCR motif enrichment based on genetic risk, offering a potential metric for autoreactivity and basis for TCR-based diagnostics and therapeutics.

Since the rise of generative AI models, many goal-directed molecule generators have been proposed as tools for discovering novel drug candidates. However, molecule generators often produce highly similar molecules and tend to overemphasize conformity to an imperfect scoring function rather than capturing the true underlying properties sought. We rectify these two shortcomings by offering diversity-based evaluations using the #Circles metric and considering constraints on scoring function calls or computation time. Our findings highlight the superior performance of SMILES-based autoregressive models in generating diverse sets of desired molecules compared to graph-based models or genetic algorithms.

A central problem in drug discovery is to identify the interactions between drug-like compounds and protein targets. Over the past few decades, various quantitative structure-activity relationship (QSAR) and proteo-chemometric (PCM) approaches have been developed to model and predict these interactions. While QSAR approaches solely utilize representations of the drug compound, PCM methods incorporate both representations of the protein target and the drug compound, enabling them to achieve above-chance predictive accuracy on previously unseen protein targets. Both QSAR and PCM approaches have recently been improved by machine learning and deep neural networks, that allow the development of drug-target interaction prediction models from measurement data. However, deep neural networks typically require large amounts of training data and cannot robustly adapt to new tasks, such as predicting interaction for unseen protein targets at inference time. In this work, we propose to use HyperNetworks to efficiently transfer information between tasks during inference and thus to accurately predict drug-target interactions on unseen protein targets. Our HyperPCM method reaches state-of-the-art performance compared to previous methods on multiple well-known benchmarks, including Davis, DUD-E, and a ChEMBL derived data set, and particularly excels at zero-shot inference involving unseen protein targets. Our method, as well as reproducible data preparation, is available at https://github.com/ml-jku/hyper-dti.