BAYESIAN OPTIMIZATION STRATEGIES IN BIOINFORMATICS: IMPROVING ACCURACY IN PROTEIN STRUCTURE PREDICTION

The accelerating global need for accurate protein structure prediction, critical for drug discovery and disease modeling, has heightened interest in Bayesian optimization strategies, particularly in low-resource environments like Ghana. This study investigates the application of Bayesian optimization techniques to improve predictive accuracy in protein structure modeling, addressing both scientific advancement and regional capacity gaps. Using a secondary data analysis approach, 105 validated protein modeling project datasets from 2020-2024 were examined, sourced from major Ghanaian research centers such as WACCBIP and KNUST. Bayesian optimization strategies, including acquisition functions, surrogate modeling techniques, and optimization parameters, were quantitatively analyzed through correlation and regression models. The major findings revealed significant positive impacts of these strategies on prediction accuracy, with the regression results indicating strong coefficients (β = 0.348 for acquisition functions, β = 0.412 for surrogate modeling techniques, β = 0.375 for optimization parameters, all p < 0.001) and an overall model fit of R² = 0.794. The correlation coefficient between surrogate modeling techniques and protein prediction accuracy was the highest (r = 0.821, p < 0.01). Conclusions confirmed that Bayesian optimization substantially improves protein structure prediction accuracy, even under computational constraints. The results imply that with targeted investments in optimization workflows, Ghana can achieve bioinformatics outcomes comparable to global benchmarks. It is recommended that bioinformatics research centers in Ghana adopt dynamic exploration-exploitation balancing, surrogate model refinements, and localized parameter tuning to maximize prediction efficiency and accuracy in resource-limited settings.