Pharmaceuticals and cosmetics are highly essential in our day-to-day life, with several benefits to human health and lifestyle. Unfortunately, these chemicals end up in different environmental compartments and thus, raise concerns for their adverse effects on non-target species including aquatic life. There are at least three stages when these pharmaceuticals and cosmetics products get released in the aquatic environment: manufacturing, usage, and disposal. Many pharmaceuticals and cosmetics being water-soluble and non-volatile can stay in the aquatic environment for a longer period, especially those that are not easily biodegradable. Thus, possess higher possibilities of showing acute as well as chronic toxicity to the aquatic life.

Acute and Chronic toxicity testing of such chemicals is thus vital but is expensive, time-consuming, and most importantly face ethical concerns due to the use of animals. And here comes the important role of computational methods as an alternative to toxicity testing on animals, along with being faster and much cheaper.

The main goal of the ‘EcoCosmePharm’ project is to develop computational models for predicting the aquatic toxicity (acute and chronic) of pharmaceuticals and cosmetics. Thus, we have developed several QSAR/QSTR models (including multi-tasking) for identifying the relationship between the structural features and properties of our interest, such as:

• Aquatic toxicity (acute and chronic)

• Biodegradation

• Bioaccumulation

Note: Please see the image carousel below illustrating all the QSAR/QSTR models developed in the project

Multi-tasking QSAR/QSTR models developed in this project are really unique (no exaggeration!!) since these models have capabilities to capture and provide much more information if compared to typical QSAR/QSTR models. The key feature of a multi-tasking model is that it not only identifies the relationship between the structural features and the biological activity/property/toxicity of our interest, but it also captures the changes in the activity/property/toxicity due to variation in the experimental conditions/parameters. Thus, while using a multi-tasking QSAR model, the user has the flexibility to get predictions according to the set experimental conditions. For instance, in this project, we have developed a multi-tasking QSTR model for the prediction of aquatic toxicity (acute and chronic) that has captured several experimental conditions such as test organisms, endpoint type, duration (in hours), media type, and exposure type. For a better explanation, we have mentioned two observations (see Table below) from the data set we have employed for the development of this multi-tasking model. As shown in the table below, each set of observations comprises the same chemical (presented in SMILES notations) tested for aquatic toxicity in different experimental conditions/parameters. The colored cells highlight the variation in the experimental parameters for that particular observation. One can easily observe that the toxicity response values vary with the changes in the experimental conditions. One might argue that the differences in the toxicity values (reported in mg/L) look small, however, note that this small difference is enough to categorize them in different toxicity categories. As we mentioned before, the developed multi-tasking QSTR model can predict the exact toxicity category/class according to the experimental conditions provided or selected by the user. Note that multi-tasking QSTR models are classification-based, thus can predict the class and not the numerical values like regression models.