Traditional smRNA-FISH probe design limits sensitivity and specificity.

Panels A–E illustrate conceptual smRNA-FISH design scenarios affecting signal-to-noise ratio (SNR). (A) Spot SNR is determined by the number of probes bound to on-target RNA (signal, S) relative to probes bound to off targets and cellular autofluorescence (background, B). On-target and off-target localization are assumed to be random. (B) A typical design shows moderate background and submaximal spot intensity due to suboptimal probe number or affinity, resulting in overlapping signal and background distributions and reduced SNR. (C) In low expression conditions, the background remains constant but fewer on-target transcripts yield fewer detectable spots, reducing overall detection sensitivity. (D) In knockout or no-expression cases, background persists while on-target signal is absent, highlighting the contribution of off-target binding and autofluorescence. (E) An optimized design minimizes background and maximizes spot intensity by using many high-affinity probes, resulting in high SNR and improved detection performance.

TrueProbes systematically evaluates on- and off-target binding dynamics to generate probe sets optimized for individual experimental contexts.

(A) Traditional smRNA-FISH design tools often select probes sequentially from the 5′ to 3′ end of the target transcript, introducing bias based on probe position rather than optimizing for binding strength or specificity. (B) Most software platforms tile the RNA with candidate probes of defined length, then apply heuristic filters to select probes, without global ranking. (C) In contrast, our approach ranks all candidate probes based on experimentally relevant specificity scores, prioritizing probes with strong on-target binding and minimal off-target interactions. (D) The pipeline begins by tiling the full target RNA sequence and generating all possible probe sequences. Each candidate is screened for off-target binding events of ≥15 nt using BLAST, assessed for RNA secondary structure, and evaluated for thermodynamic stability, including hairpins, cross-dimers, and rRNA interactions. Probes that bind rRNA are excluded. (E) Among probes with no detectable off targets, the software selects those that provide maximum coverage of the target RNA, ranking candidates by corrected on-target binding affinity after accounting for RNA secondary structure. (F) For probes with off targets, with or without incorporating cell-type–specific expression data, the remaining candidates are ranked by their specificity (low off-target binding, strong on-target binding, and minimal self-structure). The final probe set is assembled by iteratively selecting top-ranked probes that do not cross-dimerize with one another.

Systematic probe design improves signal and reduces off-target background.

(A) Absolute number of probes designed for ARF4 (left) and percentage of gene coverage (right) for each software: TrueProbes (green), Stellaris (blue), Oligostan-HT (pink), PaintSHOP (cyan), and MERFISH (orange). (B–D) Cumulative number of probes predicted to bind off-target RNAs at equilibrium under three transcript expression assumptions: equal expression across genes (B), average cell line expression (C), and Jurkat-specific transcript expression (D). (E–G) Ratio of off-target to on-target probe binding events under the same conditions, quantifying specificity of each design. Panels highlight the advantage of expression-aware, thermodynamically optimized probe design in minimizing background and maximizing true signal.

Experimental validation of smRNA-FISH probe designs for ARF4.

Representative Jurkat cell field of view with ARF4 RNA spots (green) and DAPI staining (blue) for no probe control (A), TrueProbes (B), Stellaris (C), Oligostan-HT (D), PaintSHOP (E), and MERFISH (F). Distributions are shown for spot signal intensity (G, L, Q, V), background intensity (H, M, R, W), signal-minus-background (I, N, S, X), signal-to-noise ratio (J, O, T, Y), and single-cell spot counts (K, P, U, Z). Each row corresponds to a different design tool: Stellaris (G–K, blue), Oligostan-HT (L–P, magenta), PaintSHOP (Q–U, cyan), and MERFISH (V–Z, beige), with all compared to TrueProbes (green) across three experimental replicates. TrueProbes: 409 cells, 1841 RNA spots; Stellaris: 348 cells, 1753 spots; Oligostan-HT: 283 cells, 1,014 spots; PaintSHOP: 291 cells, 191 spots; MERFISH: 327 cells, 314 spots.