RNA-seq as a screening readout: DRUG-seq vs. classic RNA-seq.

RNA sequencing (RNA-seq) has evolved from a specialised research technique into a core component of drug discovery workflows. As screening strategies increasingly move beyond single-endpoint phenotypes (such as viability, reporter assays, or cell morphology) transcriptional profiling offers a scalable and information-rich readout of cellular state.

Rather than measuring a single pathway or biochemical output, transcriptomics captures coordinated gene expression changes across thousands of genes simultaneously, enabling assessment of mechanism-of-action (MoA), target engagement, and pathway modulation in early discovery.

The challenge is that traditional RNA-seq was designed for depth and completeness rather than throughput. It delivers high sensitivity, broad transcript representation, and exceptional dynamic range, but it is also costly and operationally intensive. When the objective is to profile hundreds or thousands of perturbations in parallel, classical RNA-seq becomes impractical as a primary screening modality. DRUG-seq (Digital RNA with pertUrbation of Genes) was first described by Ye et al. in 20181. Its central innovation is early barcoding and sample pooling: each well receives a unique molecular barcode during reverse transcription, allowing thousands of samples to be pooled prior to amplification and sequenced together. This reduces per-condition cost to the low single-digit range. Importantly, the protocol begins with direct cell lysis, eliminating the need for extraction1.

This method shifts transcriptomics from a secondary validation tool into a scalable first-line screening readout. The table below provides a direct comparison across key parameters.

Table 1. Head-to-head comparison of classical RNA-seq vs DRUG-Seq.

Technical performance and what changes at scale

DRUG-seq focuses sequencing on the 3’ end of polyadenylated transcripts, reducing library complexity while preserving gene-level quantification. For perturbation analysis, where the goal is differential expression and signature clustering rather than isoform discovery, this design concentrates sequencing effort on the genes most informative for MoA inference.

Sequencing depth is typically 0.25–2 million reads per well, versus 20–50 million for classical RNA-seq. At these shallow depths, DRUG-seq detects >7,000 to ~12,000+ per sample and recovers most gene-level perturbation signatures observable with conventional RNA-seq, while remaining less sensitive for low-abundancy transcripts^1,2. Per-condition cost ranges from approximately $3–10 including triplicates, a ~100-fold reduction. For a 384-well plate, total preparation and sequencing cost approaches that of a single classical RNA-seq library.

Published benchmarking supports the method’s reliability as a screening platform. In the original study, DRUG-seq profiled 433 compounds across 8 dose levels, with compounds clustering cleanly by pharmacological class (including proteasome inhibitors, topoisomerase inhibitors, HDAC inhibitors, and kinase inhibitors) at shallow read depths¹. Technical replicate correlations range from R ≈ 0.90–0.95 and biological replicates from R ≈ 0.85–0.92, comparable to L1000 and other established perturbation profiling platforms^1,3. Published implementations also demonstrate profiling at the scale of thousands of compound–dose conditions with ~5 days turnaround².

Early barcoding also reduces handling variability and batch effects, which becomes increasingly important when processing tens of thousands of wells across multiple days or operators.

Combining Cell Painting and DRUG-Seq

One of the relevant features of DRUG-seq is its compatibility with prior Cell Painting acquisition from the same well. Imaging and transcriptomics capture overlapping but non-identical aspects of cellular state. Morphology reflects phenotypic consequences of perturbation, while transcriptomics captures the underlying regulatory response. Used together, both approaches can improve MoA inference and help distinguish compounds that appear similar in one assay but diverge in the other. For broader context on Cell Painting perturbation datasets at scale, see the JUMP Cell Painting Consortium preprint⁴.

Conclusion

DRUG-seq is most appropriate for large-scale campaigns involving hundreds to tens of thousands of compound conditions, where primary goals include compound clustering, MoA assignment, target deconvolution, transcriptional dose-response profiling, or chemical series optimisation using pathway-level readouts. Classical RNA-seq remains the better choice for full-length transcript characterization, splice isoform analysis, and more sensitive detection of low-abundance transcripts. In practical terms, DRUG-seq and classic RNA-seq are complementary technologies: DRUG-seq for scale and early triage, classical RNA-seq for depth and validation.