A workflow combining single-cell CRISPRi screening and a supervised autoencoder neural network to detect subtle transcriptomic perturbations induced by lncRNA Knock-Down

Recent advances in cancer genomics have highlighted aberrant expression of various families of non-coding RNAs in all cancer types, including lung adenocarcinomas (LUAD). Here we aim to better understand the functions of long non coding RNAs (lncRNAs) regulated by the hypoxic response in LUAD cells, conditions that promote tumor aggressiveness and drug resistance. We performed a single-cell CRISPR-interference-based (CRISPRi) transcriptome screening (CROP-Seq) for HIF1A, HIF2, and a subset of lncRNA candidates regulated by hypoxia and/or potentially associated with LUAD prognosis. The mini-CROP-seq library of validated guides RNA (gRNA) was amplified and transduced in A549 LUAD cells cultured in normoxia or exposed to hypoxic conditions during 3, 6 or 24 hours. To overcome the challenge of detecting subtle gRNA-induced transcriptomic perturbation and classifying the most responsive cells, we used a new supervised autoencoding neural networks method (SAE), leveraging on both transcriptomic data and cell labels corresponding to known received gRNA. We first validated the SAE approach on HIF1A and HIF2 by confirming the specific effect of their knock-down during the temporal switch of the hypoxic response. Next, the SAE method was able to detect stable short hypoxia-dependent transcriptomic signatures induced by the knock-down of some lncRNA candidates, outperforming previously published machine learning approaches. This proof of concept demonstrates the relevance of the SAE approach for deciphering weak perturbations in single-cell transcriptomic data readout as part of CRISPR-based screening. ### Competing Interest Statement The authors have declared no competing interest.