Single-cell population level data integration

Excited to share our new work Population-level integration of single-cell datasets enables multi-scale analysis across samples . We present scPoli to learn representation for cell types and samples that can be updated with new cells and samples.

New preprint on integrating T-cell receptor sequences and phenotype (scRNA-seq)

Proud to present mvTCR, integrating T-cell receptor sequences (α/β chain learned via transformers) and phenotype (scRNA-seq) jointly measured (e.g., Immune Profiling ) to assemble (and continually update) multi-omic T-cell atlases.

scArches wins MDSI Best Paper of the Year Award

Proud that our paper on life-long and transfer learning for single-cell biology has been awarded among top 3 papers selected by anonymous reviewers from MDSI at Technical University of Munich. This was among the work done during my doctoral studies at Helmholtz Munich and Life science school at TUM.

New paper accepted at NeurIPS 2022

Our paper titled Predicting single-cell perturbation responses for unseen drugs is now accepted as at NeurIPS 2022. In this we add chemical graph encoder to my previous work CPA to predict gene expression response to unseen drugs not observed during training.

scArches now can map scRNAseq data to spatial references

I am excited to announce the first spatial reference mapping in scArches powered by SageNet to map dissociated single cells scRNAseq into a common coordinate framework using one or more spatially resolved reference datasets.

treeArches for learning cell-type hierarchies

I am excited to share our new work, we tackled: 1) how to learn a harmonized cell-type hierarchy/taxonomy across many studies with different annotations, 2) how to automatically identify novel cell states (e.

Spotlight at Machine Learning for Drug Discovery at ICLR 2022

Our paper titled Predicting single-cell perturbation responses for unseen drugs is accepted as spotlight talk at MLDD 2022. This paper augments my previous work CPA with a molecular representation to predict unseen drugs.

New preprint on biologically informed deep learning to infer gene program

Excited to share our new approach, expiMap, to learn gene programs (GP) activity from single-cells “biologically informed deep learning”.We add prior knowledge while learning new cellular circuits, going beyond data integration and towards interpretability.

scArches is on the cover of Nature Biotechnology

Great news about scArches, we have made it to the January issue of Nature Biotechnology.

Compositional Perturbation Autoencoders (CPA) in colab with FACEBOOK AI

I am excited to announce my preprint titled Compositional perturbation autoencoder for single-cell response modeling is not out! This is the work from my time at FACEBOOK AI. The preprint is available on biorxiv.