Figure 2: ProCapNet contribution scores highlight a refined set of canonical promoter motifs.

(A) Recurrent high-scoring sequence features predictive of initiation identified by TF-MODISCO. PWM: Position-Weight Matrix; CWM: Contribution-Weight Matrix (base frequencies weighted by contribution scores). Columns 1–2 show motifs at normalized heights for visual clarity, while CWM weight (column 3) indicates the magnitude of the motif’s overall contribution, equal to the actual y-axis scale of the CWM. PWMs, CWMs, and CWM weights shown here are from the profile task’s TF-MODISCO output. Average profiles are centered on motif instances, aligned to the same orientation as the PWM/CWM. (B) Identified motif instances relative to PRO-cap peak summits. Uni., unidirectional; Bi., bidirectional (divergent transcription); Pos., positive-strand. Red vs. blue indicates motif orientation, and gray lines indicate PRO-cap summits (omitted from CA-Inr plot because it overlaps the motif hits). (C) Subpatterns found by TF-MODISCO for the NRF1 and ETS motifs that suggest Inr-like secondary roles.