Among the 33, 52 and 31 domains present in the
selected CASP8, CASP9 and CASP10 TBM-HA targets with atleast 1 TBM-HA domain, our developed
algorithm has predicted models with improved GDT-TS score for 31, 43 and 25
domains in comparison to their best structures predicted during the CASP. Overall for all the selected CASP target
domains, our predicted models have shown an average respective GDT-TS, GDT-HA
and TM_Score improvement of 3.531, 4.814 and 0.022 over the best structures
predicted during the CASP. For
most of these targets, my models are even more accurate than the most accurate
structures predicted during the CASP test. My
modelling accuracy ranks among the top10 TBM algorithms of the CASP for each of
the considered target sequences.
We developed and employed the following tactics briefly in our algorithm
- Improved template combination and modelling algorithm: In this study, I have modelled all the CASP8, CASP9 and CASP10 targets that encodes at-least 1 TBM-HA domain. Here through the reliable scores, pairwise and MSA alignment of the templates are evaluated against a target sequence for statistically ranking them through their best possible derived score equation in plausibly the order of their modelling accuracy for the target. The best template set computed by my automated C and python programmes is then employed to model the target sequence.
- Improved Protein Model Ranking through Topological Assessment: Here, I have found the logical lacuna of the all the model assessment measures including TM_Score, GDT and SphereGrinder scores. Complementarily synchronous sampling undulations are used to fix the best set of assessment measures to select the best possible structure among the sampled set of model decoys.
- Improved Protein Model Sampling Accuracy: Here, I employed the best feasible assessment measures to sample the energetic landscape of a protein sequence. The target protein models are iteratively sampled in this step until the employed scoring set converges for atleast 3 consecutive runs. Optimally sampled target model is thus lastly selected as the best predicted target model.