Off-target liability rating for — most potent siRNAs selected from — siRNAs

Examine on- and off-target activity of individual siRNAs

Select a point in either plot to inspect the siRNA.

On-target activity

Loading predictions…

Off-target profile: High-Confidence Off-Target Predictions vs String-Matching

Loading predictions…

Cutoffs

High Liability≥5 high-confidence NPMs, >100 high-confidence seed matches, or Off-Target Burden ≥ 0.08
Moderate Liability1–4 high-confidence NPMs or 10–100 high-confidence seed matches, and Off-Target Burden < 0.08
Low LiabilityNo high-confidence NPMs, 2–9 high-confidence seed matches, and Off-Target Burden < 0.08
Negligible LiabilityNo high-confidence NPMs, ≤1 high-confidence seed matches, and Off-Target Burden < 0.08

* Based on a virtual screen using DeepRNAi predictions. See Validation and Details.

Validation and Details

Sequence MatchingString-matching methods count 1) off-target transcripts with a near-perfect match (NPM) anywhere in the mature transcript allowing up to two mismatches across guide nucleotides 2–16 and 2) seed matches (7mer-m8, 7mer-A1 or 8mer) in the 3′ UTR.

High-Confidence Off-Target PredictionsDeepRNAi predictions at our High-Confidence threshold had a 65% experimental validation rate in RNA-seq.

Off-Target BurdenOff-Target Burden measures the global seed and NPM knockdown activity of an siRNA. When comparing our RNA-seq data for hepatotoxic compounds from the Janas 2018 study (“Bad Actors”) to five FDA-approved siRNA drugs (“Good Actors”), we saw a difference in the overall Off-Target Burden that could distinguish these two classes. This separation in Off-Target Burden was well-predicted by DeepRNAi.

RNA-seq and DeepRNAi Off-Target Burden for toxic compounds and clinical siRNAs. RNA-seq: Mann–Whitney U p = 0.0051, AUROC = 0.94. DeepRNAi: Mann–Whitney U p = 0.0088, AUROC = 0.91.
Not all FDA-approved siRNA drugs (Good Actors) were analyzed, due to oligo availability at the time of screening.

DeepRNAi Predictions Approach RNA-seq AccuracyThe same unseen siRNAs were tested independently in HepG2 and Huh7. We compared agreement (Pearson correlation) between these experiments to DeepRNAi predictions for NPMs and seed matches. The experiments in different cell lines were compared to each other, and DeepRNAi predictions were compared to the mean RNA-seq response across two cell lines.

The same unseen siRNAs are measured independently by RNA-seq in HepG2 and Huh7. DeepRNAi predictions for seed and NPM matches are compared with the mean of HepG2 and Huh7 RNA-seq; experimental reproducibility is assessed between the two cell lines.
Pearson correlations for NPMs and seed matches: DeepRNAi 0.75 and 0.56; RNA-seq 0.80 and 0.54. All four bars have separate legend keys.

Case StudyIn a virtual screen of 3,615 siRNAs for potency and Off-Target Burden for siRNAs targeting PCSK9, DeepRNAi ranks Inclisiran as one of the most potent compounds with low Off-Target Burden.

Predicted PCSK9 log₂FC versus predicted Off-Target Burden, with Inclisiran and the top candidates highlighted.

Design setEvery siRNA for the target transcript was scored for on-target potency, and the 100 most potent were retained for off-target profiling.

DisclaimerA Low Liability or Negligible Liability rating does not establish safety. Similarly, a Moderate Liability or High Liability rating does not imply toxicity. As per FDA guidelines, all candidates require experimental follow-up.