Posters - Schedules

Posters Home

Breast cancer is one of the leading causes of premature death for women across the world. Particularly severe cases are characterised by the appearance of metastases in addition to the primary tumour. Despite the prevalence of this disease, however, its dynamics are to this day poorly understood, to a large part due to the high degree of variation in disease progression and patient outcomes. We aim to shed more light on the behaviour of this disease by analysing data from a large-scale register of patients with metastasised breast cancer across several care centres in Austria.

We propose an approach of modelling the disease progression of patients by using a Markov Model, specifically a continuous-time Markov Chain whose states are different configurations of metastasis sites. To estimate transition rates, we use the maximum-likelihood method, directly evaluating the likelihood function using the matrix exponential.

Preliminary results show that known clinical ground truths, such as the increased aggressiveness of triple-negative breast cancer (TNBC), can be successfully recovered using this approach. Moreover, they suggest that models perform markedly better when including interaction terms between metastasis sites, which may lead to important novel insights into the underlying biological processes.

It is a vital step to evaluate drug-like compounds in terms of absorption, distribution, metabolism, excretion, and toxicity (ADMET) in drug design. Classical single-task learning based on abundant labels has achieved inspiring progress in predicting individual ADMET endpoints. Multi-task learning, having the low requirement of endpoint labels, can predict multiple ADMET endpoints simultaneously. Nonetheless, it is still an ongoing issue that the existing MTL-based approaches depend on how appropriate participating tasks are. Furthermore, there is a need to elucidate what substructures are crucial to specific ADMET endpoints. To address these issues, this work constructs a Multi-Task Graph Learning framework for predicting multiple ADMET properties of drug-like small molecules (MTGL-ADMET) under a new paradigm of MTL, ‘one primary, multiple auxiliaries’. It first leverages the status theory and the maximum flow to select appropriate auxiliary tasks of a specific ADMET endpoint task. Then, it designs a novel primary-centered multi-task learning model. The comparison with state-of-the-art MTL-based methods demonstrates the superiority of MTGL-ADMET. More elaborate experiments validate the contributions of each module. Furthermore, it illustrates the interpretability of MTGL-ADMET by indicating crucial substructures w.r.t. the primary task. It’s anticipated that this work can boost pharmacokinetic and toxicity analysis in drug discovery.

Finding molecules with desired pharmaceutical properties is crucial in drug discovery. Generative models can be an efficient tool to find desired molecules through the distribution learned by the model to approximate given training data. Existing generative models (i) do not consider backbone structures (scaffolds), resulting in inefficiency or (ii) need prior patterns for scaffolds, causing bias. Scaffolds are reasonable to use, and it is imperative to design a generative model without any prior scaffold patterns.
We propose a generative model-based molecule generator, Sc2Mol, without any prior scaffold patterns. Sc2Mol uses SMILES strings for molecules. It consists of two steps: scaffold generation and scaffold decoration, which are carried out by a variational autoencoder and a transformer, respectively. The two steps are powerful for implementing random molecule generation and scaffold optimization. Our empirical evaluation using drug-like molecule datasets confirmed the success of our model in distribution learning and molecule optimization. Also, our model could automatically learn the rules to transform coarse scaffolds into sophisticated drug candidates. These rules were consistent with those for current lead optimization.