cv

Research Scientist

Isomorphic Labs, London, UK

Postdoctoral Fellow

Mila - Quebec AI Institute and Université de Montréal, QC, Canada

• Geometric deep learning and graph signal processing; particularly studying the problem of long-range interactions in graph data.
• Supervisors: Guy Wolf (U de Montréal) and William L. Hamilton (McGill U)
• Main projects:
  • GraphGPS [NeurIPS 2022]: Developed a methodology and framework for building general, powerful, scalable (GPS) graph Transformers with linear complexity for graph representation learning and property prediction. This approach combined 3 main ingredients: (i) graph positional/structural encoding, (ii) local message-passing mechanism, and (iii) global attention mechanism. At the time of publishing GPS reached SOTA performance on 7 from 11 tested benchmarks.
    -- An extended version, GPS++ developed in collaboration with Graphcore, won the first place in OGB-LSC 2022 PCQM4Mv2 challenge.
  • Taxonomy of graph datasets [LoG 2022]: Many different types of GNN models have been developed, with many benchmarking procedures to demonstrate the superiority of one GNN model over the others, however there is a lack of systematic understanding of the underlying benchmarking datasets, and what aspects of the model are being tested. We provide a principled approach to taxonomize graph benchmarking datasets by carefully designing a collection of graph perturbations to probe the essential data characteristics that GNN models leverage to perform predictions.
  • HGNet [IEEE MLSP 2021, Best Paper Award]: Developed HGNet method, which for any two connected nodes guarantees existence of message-passing paths of at most logarithmic length w.r.t. the input graph size. Yet, under mild assumptions, its internal hierarchy maintains asymptotic size equivalent to that of the input graph. This construction outperforms conventional stacking of GCN layers particularly in tasks that require long-range interactions and in molecular property prediction benchmarks.

Software Engineering Intern

Google, Mountain View, CA, USA

• Developed machine learning models for predicting drug effects on cell phenotype from the drug molecular structure. Work done in Google Accelerated Science team.
• SMILES variational autoencoders, convolutional nets, graph neural nets
• TensorFlow, Python

Bioinformatics Research Intern

Atomwise, San Francisco, CA, USA

• Developed 3D convolutional neural nets to predict small molecule binding to protein targets and for scoring of the 3D binding poses.
• Torch, Python, R

PhD in Computer Science

University of Toronto, Canada

• Dissertation: Latent-variable models for drug response prediction and genetic testing
• Supervisors: Anna Goldenberg and Michael Brudno (co-supervisor)
• Main projects:
  • DrVAE: Personalized drug treatment outcome prediction can increase chance of recovery for cancer patients. I developed Dr.VAE, a deep generative model based on variational autoencoders, that outperforms standard classification methods for 23 out of 26 tested FDA-approved drugs. It is the first approach to leverage modelling of drug effects on gene expression to improve the final predictions.
  • Domain adaptation from lab experiments to patients: Analysed and tested transfer of ML models trained on preclinical data to patients and identified obstacles in the way.
  • fCNV: Developed a Hidden Markov model for detecting major genomic defects in a fetus during the pregnancy. From sequenced cell-free DNA collected from the pregnant mother's plazma during the pregnancy it is possible to detect most of the major symptomatic copy number alternations.

MSc in Computer Science

Comenius University in Bratislava, Slovakia

• Master’s thesis: Computational Complexity and Practical Implementation of RNA Motif Search
• Supervisor: Bronislava Brejová

BSc in Computer Science

Comenius University in Bratislava, Slovakia

• Bachelor’s thesis: RNA Motif Search in Genomic Sequences
• Supervisor: Tomáš Vinař