Unlocking the power of AI models: exploring protein folding prediction through comparative analysis

Area of Science:

• Computational biology
• Structural biology
• Bioinformatics

Background:

• Deep learning models advance protein structure prediction from sequences.
• Accurate prediction is vital for proteins with unknown structures, rare or complex forms.
• Antimony resistance marker (ARM) proteins (ARM58, ARM56) in Leishmania spp. have domains of unknown function.

Purpose of the Study:

• To evaluate the accuracy of deep learning model predictions for ARM58 and ARM56 proteins.
• To analyze prediction reliability metrics and understand model output diversity.
• To compare predictions with homologous proteins in other species, like Trypanosoma.

Main Methods:

• Utilized deep learning models for protein structure prediction.
• Assessed prediction reliability using various metrics.
• Compared predicted structures of ARM58 and ARM56 with orthologs in Trypanosoma cruzi and Trypanosoma brucei.

Main Results:

• Model predictions for ARM58 and ARM56 were assessed for accuracy and reliability.
• Analysis provided insights into the complexities and supporting metrics of the predictions.
• Comparisons were made between predictions for proteins from different species.

Conclusions:

• Evaluating diverse deep learning model outputs is essential for protein structure determination.
• Cross-species comparisons enhance understanding of protein structure and function.
• This study highlights the importance of assessing model accuracy for novel protein structures.