A continuous in silico learning strategy to identify safety liabilities in compounds used in the leather and textile industry

Abstract

There is a widely recognized need to reduce human activity's impact on the environment. Many industries of the leather and textile sector (LTI), being aware of producing a significant amount of residues (Keßler et al. 2021; Liu et al. 2021), are adopting measures to reduce the impact of their processes on the environment, starting with a more comprehensive characterization of the chemical risk associated with the substances commonly used in LTI. The present work contributes to these efforts by compiling and toxicologically annotating the substances used in LTI, supporting a continuous learning strategy for characterizing their chemical safety. This strategy combines data collection from public sources, experimental methods and in silico predictions for characterizing four different endpoints: CMR, ED, PBT, and vPvB. We present the results of a prospective validation exercise in which we confirm that in silico methods can produce reasonably good hazard estimations and fill knowledge gaps in the LTI chemical space. The proposed protocol can speed the process and optimize the use of resources including the lives of experimental animals, contributing to identifying potentially harmful substances and their possible replacement by safer alternatives, thus reducing the environmental footprint and impact on human health.

Keywords: Computational toxicology; In silico; Leather and textile industry; Machine learning; QSAR; Read across.

Fig. 1

Schema of the system developed to compile toxicological information for substances used in LTI

Fig. 2

Changes in CMR annotations after prediction. In the left column, we have the annotations in CII. In the right column, we have the annotations in CII plus the predictions with and without the cutoff of 80% confidence. YES + Pending annotations increased from 1401 to 2092: 572 predictions under 80% of confidence and 119 higher than 80%. NO annotations increased from 477 to 982: 502 predictions under 80% of confidence and 3 (cannot be seen in the figure) above 80%. No Information annotations have decreased from 2756 to 1560 if all predictions are taken, but only to 2634 if the predictions higher than 80% are considered. The amount of non-informed compounds filled is 43.4%, but for high-confidence predictions only 4.43%

Fig. 3

Changes in PBT annotations after prediction. In the left column we have the annotations in CII. In the right column we have the annotations in CII plus the predictions with and without the cutoff of 80% confidence. YES annotations increased from 38 to 123: 85 predictions under 80% of confidence, but none higher than 80%. NO annotations increased from 357 to 2932: 2528 predictions under 80% of confidence and 47 above 80%. No information annotations decreased from 4239 to 1579 if all predictions are taken, but only to 4192 if the predictions higher than 80% are considered. The amount of non-informed compounds filled is 62.75%, but for high-confidence predictions only 1.11%

Fig. 4

Changes in vPvB annotations after prediction. In the left column, we have the annotations in CII. In the right column, we have the annotations in CII plus the predictions with and without the cutoff of 80% confidence. YES annotations increased from 43 to 115: 69 predictions under 80% of confidence and 3 (cannot be seen here) higher than 80%. NO annotations increased from 357 to 2912: 2.526 predictions under 80% of confidence and 29 above 80%. No Information annotations decreased from 4234 to 1607 if all predictions are taken, but only to 4202 if the predictions higher than 80% are considered. The amount of non-informed compounds filled is 62.04%, but for high-confidence predictions only 0.75%

Fig. 5

An example of the RAX result for the ED endpoint. In the left column, a set of structures of positive compounds in CII is shown, and in the middle column, a set of non-informed compounds in CII that are very similar to the positive ones is depicted. In the right column, we can see the value of the similarity between each pair of compounds (Tanimoto index, computed using RDKit fingerprints, as described in the Materials and methods section)