Enabling Single‐Cell Drug Response Annotations from Bulk RNA‐Seq Using SCAD

Abstract

The single-cell RNA sequencing (scRNA-seq) quantifies the gene expression of individual cells, while the bulk RNA sequencing (bulk RNA-seq) characterizes the mixed transcriptome of cells. The inference of drug sensitivities for individual cells can provide new insights to understand the mechanism of anti-cancer response heterogeneity and drug resistance at the cellular resolution. However, pharmacogenomic information related to their corresponding scRNA-Seq is often limited. Therefore, a transfer learning model is proposed to infer the drug sensitivities at single-cell level. This framework learns bulk transcriptome profiles and pharmacogenomics information from population cell lines in a large public dataset and transfers the knowledge to infer drug efficacy of individual cells. The results suggest that it is suitable to learn knowledge from pre-clinical cell lines to infer pre-existing cell subpopulations with different drug sensitivities prior to drug exposure. In addition, the model offers a new perspective on drug combinations. It is observed that drug-resistant subpopulation can be sensitive to other drugs (e.g., a subset of JHU006 is Vorinostat-resistant while Gefitinib-sensitive); such finding corroborates the previously reported drug combination (Gefitinib + Vorinostat) strategy in several cancer types. The identified drug sensitivity biomarkers reveal insights into the tumor heterogeneity and treatment at cellular resolution.

MetaSCDrug: Meta-Transfer Learning for Single-Cell-Level Drug Response Prediction from Transcriptome and Molecular Representations

“…While there have been efforts to leverage deep learning networks for SC-level drug response prediction by translating knowledge from drug-cell line interactions, the optimization of these model structures tends to be drug-specific [19], e.g. scDEALL [12] and SCAD [13]. These methods require substantial amounts of target data and extensive parameter tuning for each drug to reach an optimal state suitable for transfer learning.…”

“…Moleculars selected for screening are mainly anti-cancer therapeutics, covering both targeted agents and cytotoxic chemotherapies [4]. Following the approaches of previous studies such as SCAD [13], GraphDRP [22] and GraTransDRP [23], we focused on 223 drugs applied to 1018 cell lines, with drug response values in terms of IC50 normalized within a range of (0,1). In addition, the GDSC bulk RNA-seq data was downloaded from https://www.cancerrxgene.org/downloads/bulk_download.…”

“…In their study, Kinker et al demonstrated that EpiSen-high and EpiSen-low cells, derived from the same cell lines, exhibit differential responses to a variety of drugs [27]. Building on this foundation [13], we adopted a threshold using the 10th percentile EpiSen score to define the sensitivity of each cell to all drugs. A detailed description of each dataset is provided in Table 1.…”

“…In this context, the potential of bulk cell lines pre-training followed by transfer learning on SC-level analysis emerges as a compelling approach to overcome these obstacles. A few recent studies, such as DREEP [11], scDEAL [12], and SCAD [13] attempt to leverage publicly available drug-cell line sensitivity profiles to predict drug response at the cellular level. However, previous studies have only used transcriptomic information to pre-train a single drug-specific model, which lack pharmacogenomic information for other drugs and hinder model generalization.…”

Inferring therapeutic vulnerability within tumors through integration of pan-cancer cell line and single-cell transcriptomic profiles

“…However, considering that genes may harbor varying predictability for response to different drugs and scRNA-seq data are notorious for their low detection rates as well as stochastic drop-outs 14,27 , it is not guaranteed that gene signatures always deliver reliable predictions of sensitivities to various drugs 28 . In comparison, SCAD and scDEAL directly tackle differences between bulk and SC data and emphasize integration of the two domains via neural network based approaches 23,29 . While data-hungry deep learning (DL) routes could benefit from large scRNA-seq data and model complex drug-gene relationships, the availability of CCL bulk data could pose continued limits against accurate parameter estimation.…”