Detecting subtle transcriptomic perturbations induced by lncRNAs knock-down in single-cell CRISPRi screening using a new sparse supervised autoencoder neural network

• DOI: 10.3389/fbinf.2024.1340339
• Publication Date: 2024-03-06T20:54:31Z
• Authors: Marin Truchi; Caroline Lacoux; Cyprien Gille; Julien Fassy; Virginie Magnone; Rafael Lopes Gonc...; Cédric Girard-Rib...; Iris Manosalva-Pena; Marine Gautier-Isola; Kevin Lebrigand; Pascal Barbry; Salvatore Spicuglia; Georges Vassaux; Roger Rezzonico; Michel Barlaud; Bernard Mari
• Keywords: Autoencoder; HIF1A; Guide RNA

Single-cell CRISPR-based transcriptome screens are potent genetic tools for concomitantly assessing the expression profiles of cells targeted by a set guides RNA (gRNA), and inferring target gene functions from observed perturbations. However, due to various limitations, this approach lacks sensitivity in detecting weak perturbations is essentially reliable when studying master regulators such as transcription factors. To overcome challenge subtle gRNA induced transcriptomic classifying most responsive cells, we developed new supervised autoencoder neural network method. Our Sparse (SSAE) provides selection both relevant features (genes) actual perturbed cells. We applied method on an in-house single-cell CRISPR-interference-based (CRISPRi) screening (CROP-Seq) focusing subset long non-coding RNAs (lncRNAs) regulated hypoxia, condition that promote tumor aggressiveness drug resistance, context lung adenocarcinoma (LUAD). The CROP-seq library validated against lncRNAs and, positive controls, HIF1A HIF2A, 2 main factors hypoxic response, was transduced A549 LUAD cultured normoxia or exposed conditions during 3, 6 24 h. first SSAE HIF2 confirming specific effect their knock-down temporal switch response. Next, able detect stable short hypoxia-dependent signatures some candidates, outperforming previously published machine learning approaches. This proof concept demonstrates relevance deciphering data readout part screening.