NTB-W10

• Date: Sunday, September 18th
• Time: 14:00 to 18:00 CEST (Slot 26)
• Format: Face-to-face
• Room: TBD

• María Rodríguez-Martínez
• Anna Niarakis
• Matteo Barberis

Interpretable machine learning & statistical modelling of the immune system in health and disease settings

The immune system is highly complex, and its malfunctioning can result in a wide number of disorders. A correct understanding of its inner workings is crucial to design optimal immunotherapies, develop new vaccines, understand the molecular basis of autoimmune diseases, etc. However, immune-related diseases pose specific challenges associated with the incomplete understanding of the underlying highly non-linear molecular and cellular interactions, and with the limited therapeutic options available based on existing drugs.

Recent years have witnessed the unstoppable development of high-throughput experimental technologies in molecular biology. We can now routinely profile thousands or even millions of single cells and millions of profiles are already publicly available. The availability of large amounts of molecular data has powered the development of many statistical and machine learning models focused on understanding the complexity of the immune system. Statistical and machine learning approaches have long been considered orthogonal approaches, the former being focused on modelling statistical relationships, and the latter on identifying hidden data patterns and correlations. However, the synergy between both approaches may greatly improve the accuracy and coverage of computational models, and help to disentangle the complex mechanisms that govern immune processes.

Attention will be given to the rising field of interpretable deep learning, which aims to overcome the black-box nature of most currently available deep learning models. Lack of understanding—inherent to black-box models—is detrimental in high-stake scenarios, such as biomedical research, where patients and clinicians need to understand the causal relationships underpinning model predictions. The alternative to black-box models, mechanistic models, allow to guide precise, testable predictions; however, it may be challenging to readily apply these models to high-throughput data.

By bringing together experts working in the fields of mechanistic, statistical, and AI modelling, this workshop aims to investigate how information about the molecular mechanisms underlying biological functions and cellular processes can be extracted from high-throughput data.

The target audience ranges from students, early and advanced researchers to professionals working in the fields of Immunology, Systems Biology, Computational Biology, Computer Science and Bioinformatics who apply or are interested in applying computational modelling techniques and machine learning approaches for studying cellular functions of the immune system.

TIME	CONTENT
14:00 – 14:10	Welcome and introduction to the workshop
	Session 1 – Statistical models Chair: María Rodríguez Martínez
14:10 – 14:40	Data-driven model for neo-antigen/T-cell presentation and recognition: the case of pancreatic cancer. Remi Monasson (Ecole Normale Supérieure, France)
14:40 – 15:10	Statistical and machine-learning analysis of adaptive immune specificity Victor Greiff (University of Oslo, Norway)
	Session 2 – High throughput data analysis methods Chair: Anna Niarakis
15:10 – 15:40	Inferring shared intratumor heterogeneity from bulk RNA-seq data without matched single-cell reference. Valentina Boeva (ETH, Switzerland)
	Break
16:00 – 16:30	Insights from topological data analysis of COVID-19 single-cell transcriptomic. Davide Cirillo (Barcelona Supercomputing Center, Spain)
	Session 3 – Machine learning Chair: Matteo Barberis
16:30 – 17:00	Tackling the complexity of (unseen) epitope-TCR predictions. Pieter Meysman (University of Antwerp, Belgium)
17:00 – 17:20	Interpretable machine learning to unravel T cell receptor binding rules. Anna Weber (IBM Research Europe, Switzerland)
	Session 4 – Discussion Chairs: María Rodríguez Martínez, Anna Niarakis, Matteo Barberis
17:20 – 17:50	Round table discussion with speakers and participants. Discussion on key challenges and open questions ofimmunological modelling moderated by the three organizers.
17:50 – 18:00	Closing remarks