Posts by Collection

research

talks

Model-Based Trajectory Inference for Single-Cell RNA Sequencing Using Deep Learning with a Mixture Prior

Published: April 26, 2021

Background

In the analysis of single-cell RNA sequencing (scRNA-seq) data, trajectory inference or pseudotime analysis methods are proposed to infer the cell developmental lineages and order cells on their pseudotime within these lineages. As scRNA-seq datasets continue to expand, there is a strong need to integrate different datasets and perform trajectory inference jointly to align cells experiencing similar dynamic changes but are from different labs or under various conditions. By a joint cell trajectory inference of our own data with other publicly available scRNA-seq datasets, we aim to obtain a complete picture of the neurogenesis in the mouse neocortex and expand our understanding of the progenitor heterogeneity.

Results

We develop a new quantitative framework to perform a joint cell lineage analysis, which uses a statistical hierarchical mixture model for the trajectory structure on the cells' low-dimensional embedding. Our method outperforms other state-of-the-art trajectory inference methods on both real and synthetic data under various trajectory topologies. We also apply it to jointly analyze two single-cell RNA sequencing datasets on the mouse neocortex, with a conditional variational auto-encoder and adjusting for batch and cell-cycle effects. The inferred developmental trajectory enables us to study the progenitor cell types and identify the subtle development of IPCs. After correcting post-estimation bias, we have detected around 1000 genes whose co-expression patterns with Eomes change significantly across embryonic days.

Conclusions

(1) Our method successfully merges cells from different datasets and learns a shared trajectory of neurogenesis, keeping biologically meaningful differences between the two datasets. (2) We identify progenitor cell types based on integrated data and can further divide the IPCs into post-mitotic and mitotic IPCs. (3) We are able to identify temporally expressed transcription regulators in cortical progenitors.

Subsample Ridge Ensembles: Equivalences and Generalized Cross-Validation

Published: July 28, 2023

More information here

Extrapolated cross-validation for randomized ensembles

Published: August 06, 2023

Ensemble methods such as bagging and random forests are ubiquitous in various Þelds, from Þnance to genomics. Despite their prevalence, the question of the e_cient tuning of ensemble parameters has received relatively little attention. This paper introduces a cross-validation method, ECV (Extrapolated Cross-Validation), for tuning the ensemble and subsample sizes in randomized ensembles. Our method builds on two primary ingredients: initial estimators for small ensemble sizes using out-of-bag errors and a novel risk extrapolation technique that leverages the structure of prediction risk decomposition. By establishing uniform consistency of our risk extrapolation technique over ensemble and subsample sizes, we show that ECV yields _-optimal (with respect to the oracle-tuned risk) ensembles for squared prediction risk. Our theory accommodates general predictors, only requires mild moment assumptions, and allows for high-dimensional regimes where the feature dimension grows with the sample size. As a practical case study, we employ ECV to predict surface protein abundances from gene expressions in single-cell multiomics using random forests under a computational constraint on the maximum ensemble size. Compared to sample-split and K-fold crossvalidation, ECV achieves higher accuracy by avoiding sample splitting. Meanwhile, its computational cost is considerably lower owing to the use of the risk extrapolation technique.

Simultaneous inference for generalized linear models with unmeasured confounders

Published: November 20, 2023