Accurate identification of RNA D modification using multiple features.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Dihydrouridine (D) is a crucial tRNA modification affecting tRNA flexibility and cancer development.
• Current sequencing methods for D modification detection are costly and time-consuming.
• Developing precise computational tools is essential for advancing research in molecular mechanisms and medical applications.

Purpose of the Study:

• To develop a novel computational predictor, iRNAD_XGBoost, for identifying potential D modification sites in tRNAs.
• To leverage multiple RNA sequence representations and machine learning for accurate prediction.
• To provide a cost-effective and efficient alternative to experimental sequencing techniques.

Main Methods:

• Utilized XGBoost algorithm with the top 30 selected features for model construction.
• Addressed data imbalance using a hybrid sampling method (SMOTEEEN).
• Employed multiple RNA sequence representations including chemical property and nucleotide density (CPND), electron-ion interaction pseudopotential (EIIP and PseEIIP), and dinucleotide composition (DNC).

Main Results:

• The optimized iRNAD_XGBoost model achieved high prediction accuracy with 97.13% sensitivity (Sn) and 97.38% specificity (Sp) in jackknife tests.
• Independent experimental validation yielded Sn of 91.67% and Sp of 94.74%.
• The model demonstrated superior generalizability and consistent prediction efficiency compared to the iRNAD method, achieving high Matthews Correlation Coefficient (MCC) scores (0.94 and 0.86).

Conclusions:

• iRNAD_XGBoost is a highly effective computational tool for identifying tRNA D modification sites.
• Key features such as CPND, EIIP/PseEIIP, and DNC are critical for D modification recognition.
• The predictor offers a promising avenue for experimental biologists to investigate tRNA D modification functions and their roles in diseases.