Fig. 4

Benchmarking the use of INTERFACE for selection of “true” predicted target mRNAs of under-characterized sRNAs. a INTERFACE data, along with position and motif information, allows filtration of a pool of sRNA regions targeted for the selection of likely functional regions. Specifically, likely functional regions are chosen as those that exhibit accessibility values within extremes chosen for benchmarking (>0.75 or <0.25), 5′- or 3′-end position, and, in rare circumstances, the presence of a YUNR motif (see Methods). These likely functional regions can be used to inform computational mRNA target predictions for sRNAs to suggest positive binding partners (INTERFACE-informed, left of table) that may not typically be chosen for experimental validation (top-ranked, right of table) due to low energetic favorability. b Regional position versus accessibility for three representative uncharacterized sRNAs selected for benchmarking of computational mRNA predictions. Dark gray regions correspond to likely functional regions, chosen for their extreme accessibilities in the 5′ or 3′ 20% end of the transcript. In the case of accessibility clustering, presence of a YUNR motif is indicated by a triange. Percentages of predicted mRNA target interactions within top 100 computational predictions that involve the likely functional regions are listed beside the corresponding region. c Positive predictive values (PPV) of computationally predicted binding partners for three representative uncharacterized sRNAs as evaluated using EMSA. Numbers within bars correspond to the ranking (out of 100) of all confirmed mRNA interaction partners for each prediction category (INTERFACE-informed v. top-ranked). For SroE, one mRNA target from the INTERFACE-informed group (ranking 46/100 in IntaRNA predictions) was confirmed via EMSA, although none of the top-ranked IntaRNA predicted targets tested (within top 12/100) exhibit sRNA binding (Supplementary Fig. 5). For SroG, INTERFACE correctly pinpoints a region implicated in many of the top predictions (Fig. 3b), further giving confidence to these as true partners, and also pinpoints lowly ranked targets via the two proposed functional regions at the 5′-end (Fig. 3b). Finally, considerable overlap between INTERFACE-informed predictions and top-ranked tested IntaRNA predictions (within top 14/100 rank) of Tpke70 led to the confirmation of mcrA as a target (Table 1). Importantly, IntaRNA correctly predicts a confirmed weak target (agaS) to interact with a region not captured by the INTERFACE-informed group