A pair-conformation-dependent scoring function for evaluating 3D RNA-protein complex structures

Computational prediction of RNA-protein complex 3D structures includes two basic steps: one is sampling possible structures and another is scoring the sampled structures to pick out the correct one. At present, constructing accurate scoring functions is still not well solved and the performances of the scoring functions usually depend on used benchmarks. Here we propose a pair-conformation-dependent scoring function, 3dRPC-Score, for 3D RNA-protein complex structure prediction by considering the nucleotide-residue pairs having the same energy if their conformations are similar, instead of the distance-only dependence of the most existing scoring functions. Benchmarking shows that 3dRPC-Score has a consistent performance in three test sets.

RNA-蛋白质复合物三维结构的计算预测包含两个基本步骤：一是对潜在可行结构进行采样，二是对采样得到的结构进行打分以筛选出正确构型。当前，精准打分函数的构建仍未得到妥善解决，且打分函数的性能通常依赖于所采用的基准测试集。为此，我们提出一种面向RNA-蛋白质复合物三维结构预测的、依赖构象对的打分函数3dRPC-Score：该函数认为若核苷酸-残基对的构象相似，则二者具有相同的能量，而非现有多数打分函数仅依赖距离的设计逻辑。基准测试结果表明，3dRPC-Score在三个测试集上均展现出稳定一致的性能。