6/n We benchmarked UNAGI against both conventional DL methods (e.g., scVI) and foundation models (scGPT, Geneformer). UNAGI outperformed them in cell-embedding generation, disease-associated marker/pathway identification, and in-silico drug perturbation.

5/n We evaluated the generalizability of UNAGI using a COVID-19 dataset. And UNAGI successfully 1) reconstructed the cellular dynamics of COVID19 and 2) re-discovered some compounds that have anti-COVID19 effects supported by existing literatures.
#covid19

4/n Thanks to @kaminskimed.bsky.social and his team, we validated one of the UNAGI's prediction, Nifedipine, a common medication to treat high blood pressure do have the similar anti-fibrotic effects as Nintedanib in the human Precision-Cut Lung Slices.
#ipf #singlecell

3/n UNAGI can conduct perturbation on the pathway and drug/compound levels. When applied to the IPF data, we identify a full spectrum of pathways drug/compounds from well established to less studied ones. e.g. We re-identified Nintedanib, the FDA approved drug for IPF, without prior knowledge.
#ipf

2/n We applied UNAGI to a time-series snRNA-seq Idiopathic Pulmonary Fibrosis (IPF) dataset and successfully reconstructed
(1) the temporal dynamics graph (panel A) to describe the IPF progression
(2) cellular dynamics (panel B and C) of individual cell tracks.
#singlecell #ipf

(4/n) SUICA runs in the whole gene space. It enhances the signals at gene levels (left) and recovers the original cell-type related signatures (right).

#SpatialTranscriptomics

(3/n) Based on Graph Total Variation and graph-based spectral analysis of embeddings generated by both the AE and the GAE, we showed that GAE offers a more expressive and informative representation, effectively capturing disentangled embeddings for ST.
#GAE #Single-cell #SpatialTranscriptomics

(1/n) Excited to share our new work accepted at ICML 2025. SUICA, a novel method to enhance the quality of spatial transcriptomics from multiple ST platforms including Visium, stereo-seq, slide-seq-V2....
#SpatialTranscriptomics
Paper at: arxiv.org/abs/2412.01124
Code at: github.com/Szym29/SUICA