Computing and Artificial Intelligence

The prediction of drug-target interactions (DTIs) using machine learning techniques together with the proper representation of compounds can speed up the time-consuming experimental work in predicting DTIs especially when a large dataset is used. Hence, in this paper, we have proposed a new molecular descriptor based on LINGO Profiles known as LINGO Profiles Fingerprint (LPFP). LPFP is used together with machine learning to predict DTIs on a ChEMBL dataset. Dimensionality reduction using Association Rule Mining (ARM) is also introduced to overcome the high dimensionality suffered by LPFP. LPFP managed to reach an equal accuracy reading to the state-of-the-art descriptor called ECFP4 (Δ0.18%), but it suffers in the time taken (Δ27 mins) due to the dimensionality problem mentioned. Hence, three new smaller size LPFPs (s = 60%, s = 70%, s = 80%) were constructed by only extracting the important fragments using ARM and then a benchmark analysis with the original LPFP and ECFP4 fingerprints was done. This study not only solved the dimensionality problem, but also managed to excel in both the accuracy and time taken when predicting DTIs. An increase in the accuracy of over 250 times faster than the original LPFP was observed after the benchmark analysis is performed. Furthermore, an accuracy of over 80% was achieved in three new activity classes that are acquired from ChEMBL, further proving the promising performance of ARM which has made it favourable for LPFPs to be used in DTI prediction and in other drug discovery problems.

Ongoing research on the temporal and spatial distribution of algae ecological data has caused intricacies entailing incomprehensible data, model overfit, and inaccurate algal bloom prediction. Relevant scholars have integrated past historical data with machine learning (ML) and deep learning (DL) approaches to forecast the advent of harmful algal blooms (HAB) following successful data-driven techniques. As potential HAB outbreaks could be predicted through time-series forecasting (TSF) to gauge future events of interest, this research aimed to holistically review field-based complexities, influencing factors, and algal growth prediction trends and analyses with or without the time-series approach. It is deemed pivotal to examine algal growth factors for useful insights into the growth of algal blooms. Multiple open issues concerning indicator types and numbers, feature selection (FS) methods, ML and DL forms, and the time series-DL integration were duly highlighted. This algal growth prediction review corresponded to various (chronologically-sequenced) past studies with the algal ecology domain established as a reference directory. As a valuable resource for beginners to internalize the algae ecological informatics research patterns and scholars to optimize current prediction techniques, this study outlined the (i) aforementioned open issues with an end-to-end (E2E) evaluation process ranging from FS to predictive model performance and (ii) potential alternatives to bridge the literature gaps.