The predictive index generated by this secondary algorithm classifies patients as either 'ACT-responders' or 'ACT-non-responders', depending on whether the index is above or below the predetermined classification threshold (median index of Training Series B).
Moreover, for the chemicals with COMaj between the Cat 2 and the Cat 1 classification thresholds (1 ≤ COMaj < 3), 7.7 % (20/261) of the animals from the RCD had mean CO scores below the Cat 2 classification threshold (mean CO < 1) and 6.5 % (17/261) had mean CO scores above the Cat 1 classification threshold (mean CO ≥ 3).
In the NCD, on the other hand, 13.2 % (65/492) of the animals had mean CO scores below the Cat 2 classification threshold (mean CO < 1) and only 1.2%% (6/492) of the animals had mean CO scores above the Cat 1 classification threshold (mean CO ≥ 3).
The mean CO scores for the chemicals classified as Cat 1 based on CO = 4, on the other hand, were clearly higher than those for the chemicals classified as Cat 2 based on COMaj ≥ 1, but the majority were still well within the Cat 2 scores, i.e. below the Cat 1 classification threshold (Fig. 1a).
The majority of these chemicals were classified based on CRMaj (Table 5: only 2.1%% in both the RCD and NCD were classified based on CCMaj only), but some of the individual animal mean CR scores were still below the Cat 2 classification threshold of 2, with the median of all values being at (NCD) or very close to (RCD) this cut-off score (Fig. 2a).
The mean CO scores of the animals tested with chemicals that were classified Cat 1 based on a CO = 4 (and COMaj < 3), IRMaj > 1.5 only or Persistence only are generally above the CO Cat 2 classification threshold and mostly below the Cat 1 classification threshold (1 ≤ mean CO < 3) for both databases (Fig. 1a).
Nevertheless, we decided to use 25 neighbors, as this allowed us to modify more freely the classification threshold to label a gene as "synaptic" (see below).
If the average output lies above the ensemble classification threshold, the compound is classified as a positive; if the average falls below the threshold, the compound is classified as a negative.
Finally, for the chemicals with COMaj or IRMaj above the Cat 1 classification threshold (COMaj ≥ 3; IRMaj > 1.5), a substantial 23.3%% (10/43) or 13.0%% (3/23) (RCD) and 12.5 % (6/48) or 11.1%% (3/27) (NCD) of the animals had mean CO or IR scores, respectively, below the Cat 1 classification threshold (all animals counted below the dark-grey box in Tables 6 and 7).
In the NCD and RCD, the majority of the animals tested with chemicals that were classified Cat 2 based on conjunctiva effects, but without enough corneal involvement to generate a classification, have mean CR scores that are above the Cat 2 classification threshold (Fig. 2a) but have mean CO, IR, and CC scores that are below the Cat 2 classification threshold (Figs. 1a, 2a, b).
