Page 98 - AJWEP-v22i3

Sonsare, et al.

partition coefficients of PFCA and PFSA anions. artificial intelligence-driven models to infer high-level
Furthermore, the study finds a relationship between the outcomes from molecular data – a concept that directly
neutral Kₒ and the neutral membrane-water partition supports our approach to predicting environmental
w
coefficient, implying that the more easily measured log partition coefficients from molecular structure.
Kₒ may be used to predict the log membrane-water A previous study reviewed the integration of
w
partition coefficient. This approach is used to assess machine learning with QSAR modeling for drug
experimental data and expand property data for PFCAs discovery and environmental assessment. It highlighted
and PFSAs with different chain lengths. 28 advancements in machine learning techniques that
The study presents two-parameter linear free enhance QSAR modeling, improving predictions of
energy relationship models that use the log K and toxicity and biological activity. The study emphasized
ow
the dimensionless Henry’s law constant (log K ) to the importance of molecular connectivity indices
aw
estimate the lipid–water partition coefficients (log as structural descriptors in QSAR modeling. It
K and log K ) of organic chemicals, addressing the discussed the challenges of predicting toxicity due to
lw
pw
current lack of experimental data and time-consuming limited experimental data and the need for accurate
estimation methods. The developed models have high models. The paper also addressed the environmental
predictive accuracy, with R values of 0.971 for log K impact of pharmaceuticals and the role of QSAR in
2
lw
and 0.953 for log K , and RMSEs of 0.375 and 0.413, assessing chemical risks. The study focused on using
32
pw
respectively. They can be integrated into the United aluminum-based electrocoagulation to remove from
States Environmental Protection Agency’s Estimation water. Response surface methodology and machine
Programs Interface Suite software to improve its capacity learning models optimize the electrochemical removal
for estimating the environmental properties of organic process. The best removal rates achieved were 88.21%
contaminants. The study assesses the usefulness experimentally and 93.87% predicted. Key parameters
29
of continuum solvation models paired with density affecting removal include pH, electrode type, initial
functional theory approaches in predicting the K (log concentration, and electrolysis time. The adaptive
ow
P) for 56 fluorinated medicinal compounds, concluding neuro-fuzzy inference system model outperformed other
that the density model produces log P values that are models in predicting experimental results. The study
33
consistent with benchmark data. It was observed that the presented 10 recommendations to improve the European
conductor-like polarizable continuum models struggle Medicines Agency’s guidance for environmental risk
with accurately predicting trends, frequently resulting assessment of pharmaceuticals. Recommendations
in incorrect sign reversals compared to benchmark include assessing antibiotic resistance risks and refining
values, while the choice of basis set had minimal impact, test proposals. The authors emphasized the need for
and the selection of atomic radii influenced geometry regular updates to incorporate new scientific knowledge.
convergence. The research proposes a new model for The study highlighted the importance of transparency
30
predicting the temperature dependence of the octanol- and emission data in risk assessments. Overall, the
air partition ratio, which is critical for understanding recommendations aimed to enhance environmental
chemical partitioning in environmental chemistry. The protection and societal benefits. 34
scientists used a large dataset of 195 compounds to Another study discusses the evolution of QSAR
create prediction equations for the internal energy of studies, emphasizing the significant impact of machine
phase transition (ΔU OA°). The study found substantial learning methods on this field. It highlights the
correlations between variables, with the best prediction integration of various machine learning techniques,
model attaining a high adjusted R value. This indicates including deep learning, to improve the prediction of
2
its usefulness in forecasting neutral organic chemical molecular activities and properties, which are crucial
partitioning behavior across different temperatures. 31 for drug discovery. The authors note the challenges
Related studies have demonstrated the utility faced in QSAR, such as data sparsity and the need
of artificial intelligence in complex biological and for robust experimental datasets, while advocating
environmental systems. For instance, artificial for collaborative efforts in model sharing among
intelligence has been used to connect molecular companies to improve predictive accuracy. Overall, the
and genotypic data to phenotypic traits in plant paper serves as a reference for modern QSAR methods
development and to monitor the environmental fate and applications propelled by machine learning
32
of pharmaceutical and personal care products in water advancements. The article introduces MetDNA, a
35
systems. These works underscore the potential of process of metabolism network-based recursive method
33

Volume 22 Issue 3 (2025) 92 doi: 10.36922/AJWEP025070041