Cancer arises from a complex and heterogeneous interaction of driver mutations that can affect hundreds of genes. The function of most of these genes, their interplay and how we can exploit cancer-specific dependencies for therapy is largely unknown. The recent development of spatial transcriptomes can further capture spatial information using RNA barcodes to distinguish different spots in the tissue. Understanding the spatial variation in gene information and its corresponding effects in terms of phenotypic changes to the cell captured in this transcriptome is crucial in understanding the heterogeneity within a tumour. While the data exists, an important question is how to integrate the morphology in histology images with molecular properties captured by spatial transcriptomics. 

To advance the use of machine learning (ML) and artificial intelligence (AI) Rushin Gindra from the Helmholtz AI young investigator group “AI for Microscopy and Computational Pathology” of Tingying Peng at Helmholtz Munich showed his poster entitled “self-supervision in spatial transcriptomics” during the poster session at the Helmholtz AI conference from June 2-3, 2022 at Maritim Hotel & Internationales Congress Center Dresden. 

In his poster he introduced a novel method on using self-supervision in spatial transcriptomics: This method predicts the gene-expression in each spatial spot using only histology images. Each node in the graph represents a barcoded spot in the tissue. The goal is to get the correlation between the phenotypic changes in the tissue image and the underlying genes. With the help of self-supervised models and neighbourhood dependencies of cancer tissues, he was able to predict the gene-expression from the tissue image itself, opening up wide possibilities to perform detailed analysis in existing data cohorts and also figure out therapeutic vulnerabilities. His method outperforms current state of the art methods for gene expression prediction in spatial transcriptomics.

Figure: Poster submitted by Rushin Gindra.

Over 70 posters were presented to the jury consisting of Xiaoxiang Zhu (DLR), Richard Bamler (DLR), Timo Dickscheid (FZJ), Guido Juckeland (HZDR)and Fabian Theis(Helmholtz Munich) from the Helmholtz AI steering board. First prize went to Sugandha Doda (TU Munich) with her poster entitled ‘SO2SAT POP – a curated benchmark data set for population estimation from space on a continental scale’, second prize to Patrick Stiller (HZDR) with his poster ‘In Situ Surrogate Model Training via Streaming’, and third prize to Rushin Gindra (Helmholtz AI) for his poster entitled ‘Locally attentive graph transformers for spatial transcriptomics’.

Image: Jury members Richard Bamler (DLR) and Guido Juckeland (HZDR) hand over the 3rd prize to Rushin Gindra (Helmholtz AI)

With 200 participants onsite and over 220 virtual participants, the Helmholtz AI conference 2022 was a successful event for listening to exciting talks, exchanging ideas and networking for the whole Helmholtz AI community.