The tRNAmod is a web-server for the prediction of transfer RNA (tRNA) modifications. Post-transcriptional modifications of tRNA plays a major role in their functionality. It provides both flexibility as well as rigidity and fine tune tRNA structure for the maximum performance. Modifications affect vast number of biological phenomenon such as: alternative folding of tRNA, gene expression, trnaslation speed and accuracy, enhance codon binding/discrimination, prevents frameshifting and maintain proper translation reading frame. Modification of wobble position (34^th) expand the tRNA ability for codon:anticodon recognitions.

We analyzed available 642 modified tRNAs of 77 different organisms (49 Eukaryotes, 17 Bacteria, 7 Archaea and 4 Virus) from MODOMICS database. It was observed that ~13% of bases are modified with ~57 different type of modifications. We created following Two Sample Logo (TSL) of 15-length sliding patterns of modfied and unmodified central residues (8^th position).

Most of modified bases are Uridine (55.85%). Thus, we also created following TSL for modified and unmodified Uridines (15 length; central 8^th position). It was observed that modified or unmodified Uridines are different in the position-specific preference of different neighbouring nucleotides.

Pseudo-uridine and Dihydro-uridine are two most commonly (contributes ~77% of uridine-derived modifications) present in tRNAs. Thus, we developed separate prediction models for these two modifications. Pseudo-uridine (Y) contains ~45% of all Uridine-derived modifications. It was observed in the following WebLogo (15 length; central 8^th position) that GTYC (in the WebLogo GUUC) pattern-based conservation present in T-loop but Pseudo-uridine also present in other loops (D and AC loop).

Dihydro-uridine (D) modification is present in ~32% of uridine-derived modifications. We observed in the following WebLogo (15 length; central 8^th position) that mostly Dihydro-uridine present in the D-loop and also present in the other loops of tRNAs.

As shown in the weblogo, it is clear that both Y and D don't contain any conserve site for their occurance. Therefore, it is important to predict them into the tRNA sequences. We have developed a Support Vector Machines (SVMs) based predictor tRNAmod using Binary patterns and structural information computed from an existing tRNAscan-SE software.

The details of applied algorithms is available HERE