welcome to the abante lab at universitat de barcelona

Our lab is dedicated to advancing the field of neuroscience by developing in silico models that harness machine learning to reveal biological insights. Our mission is to not only contribute to foundational scientific knowledge but also to promote a future with reduced reliance on animal experimentation. To achieve these goals, we work with multimodal omic data (e.g., scRNA-seq) as well as functional data (e.g., Ca imaging). At the core of our research, we are actively developing computational approaches that leverage the latest advances in deep generative modeling and artificial intelligence.

research areas

Our current projects encompass a range of topics at the intersection of machine learning and neuroscience. Here are two of the primary areas we are focused on:

genotype-phenotype prediction

Fig. Graph neural network approach for genotype-phenotype prediction.

In this project, we aim to improve the prediction of age of onset for Huntington’s disease by identifying genetic modifiers that current methods struggle to detect. To do this, we employ graph neural networks in combination with protein interaction networks. By training these models on SNP data from the Enroll-HD study, we can identify protein subnetworks that provide valuable insights into the genetic factors influencing Huntington’s disease onset. In addition, we are in the process of augmenting this model with other modalities predicted using genomic Large Language Models (gLLMs), such as tissue-specific gene expression prediction, which will provide the model with critical information and hopefully improve its predictive performance.

deep generative models for calcium imaging

Fig. Single-neuron deep generative models for calcium imaging data.

Our lab is actively developing deep generative models that are tailored to calcium imaging data, focusing on both single-neuron and population-level analyses: snDGM and popDGM, respectively. We are building these models such that they isolate and interpret specific sources of variability relevant to neural activity. On the one hand, the snDGM provides a foundation for creating single-cell multimodal models that will eventually integrate both transcriptomic and functional data, enabling a richer and more comprehensive understanding of neural dynamics. On the other hand, the popDGM allows us to study network dynamics as a function of other potentially time-varying covariates.

Our lab’s research pushes the boundaries of what can be achieved with in silico models in the life sciences, and we are excited to be at the forefront of discoveries that have the potential to transform neuroscience. We welcome collaborations and discussions with others interested in our work.