An inverse-task benchmark

Among the 60 methods, 14 report a comparison with another method in this set. Each of those comparisons is that method’s own validation. 9 of the 14 compare across C0–C3 classes. 125 of 135 catalogued dataset and resource identifiers are linked to a single method, so most studies do not yet inherit a shared task. Identifier reuse is summarized on the datasets page, and code deposits are on the reproducibility page.

Comparisons inside each method paper

Each row is taken from that method’s own paper. “Across C0–C3 classes” means at least one method in the Compared against column is in a different class. SiPER and DECCODE are both Class C2, so that pair is same-class. Method names open the method page. pbn-STAC opens its ScienceDirect article.

The 11 methods those 14 papers compare against

There are 11 rows here, not 14. The papers above often name the same methods. The count is how many of those papers name that method. One paper can name several methods, so the counts add up to more than 14.

Later studies

These comparisons were published after the method papers, in studies that look at several methods together. A comparison inside a method paper is that method’s own validation, so it is not an independent test of the method.

Plaugher and Murrugarra 2023

Plaugher and Murrugarra compared six Boolean control methods on one T-LGL leukemia Boolean network, with apoptosis as the objective. Control sets differ. Feedback vertex set control supplies an upper bound on the number of targets rather than a shared intervention. Every method below is Class C3. None of the six reports this comparison in its own paper.

Sordo et al. is the name used here for Sordo Vieira, Laubenbacher, and Murrugarra (2019). FVS is Mochizuki et al. 2013. Zañudo et al. 2017 is the separate structure-based feedback-vertex-set paper. Plaugher and Murrugarra cite both as feedback vertex set control.

Lauber and List 2026 (DiReG)

Lauber and List set these six transcription-factor discovery methods side by side. For each method they record the data it reads, whether code is public, the algorithm, and the strengths and limits of that design. CellNet, D'Alessio et al., and Mogrify read expression only. IRENE, Taiji-reprogram, and ANANSE also read ChIP-seq and other epigenomic data. Their web tool, DiReG, then places precomputed predictions from these methods on one screen so a user can inspect them together.