Extrapolating heterogeneous time-series gene expression data using Sagittarius

Abstract

Understanding the temporal dynamics of gene expression is crucial for developmental biology, tumor biology, and biogerontology. However, some timepoints remain challenging to measure in the lab, particularly during very early or very late stages of a biological process. Here we propose Sagittarius, a transformer-based model that can accurately simulate gene expression profiles at timepoints outside of the range of times measured in the lab. The key idea behind Sagittarius is to learn a shared reference space for time series measurements, thereby explicitly modeling unaligned timepoints and conditional batch effects between time series, and making the model widely applicable to diverse biological settings. We show Sagittarius's promising performance when extrapolating mammalian developmental gene expression, simulating druginduced expression at unmeasured dose and treatment times, and augmenting datasets to accurately predict drug sensitivity. We also used Sagittarius to extrapolate mutation profiles for early-stage cancer patients, which enabled us to discover a gene set connected to the Hedgehog signaling pathway that may be related to tumorigenesis in sarcoma patients, including PTCH1, ARID2, and MYCBP2. By augmenting experimental temporal datasets with crucial but difficultto-measure extrapolated datapoints, Sagittarius enables deeper insights into the temporal dynamics of heterogeneous transcriptomic processes and can be broadly applied to biological time series extrapolation..

scNODE : Generative Model for Temporal Single Cell Transcriptomic Data Prediction

“…Single-cell gene expression prediction has been investigated in previous studies (Cao et al, 2021) Recently, a few generative methods have introduced procedures to understand cell differentiation based on physical time and predict unobserved timepoints. Sagittarius method (Woicik et al, 2023) predicts gene expressions at future timepoints but mainly focuses on bulk RNA-seq data that requires cell matching between timepoints. This matching is hard to obtain for real-world scRNA-seq as the cells are lysed during the experiment.…”

“…Especially when extrapolating five timepoints, scNODE significantly outperforms the other two methods in DR and SC datasets, which implies that scNODE better generalizes to unobserved future timepoints that have shifted distributions. Therefore, although accurate extrapolation is a challenge (Woicik et al, 2023) in single-cell gene expression prediction, scNODE still demonstrates significant improvement over state-of-the-art methods in accurately extrapolating future timepoints.…”

“…Hence, they do not accurately model how cells develop in real physical time. Sagittarius method (Woicik et al, 2023) predicts gene expressions at future timepoints but requires cell matching between timepoints. This matching is hard to obtain for real-world scRNA-seq as the cells are lysed during the experiment.…”