在蛋白質交互作用的接觸面上有一個區塊，這區塊關係到整個蛋白質交互作用的能量鍵結，稱作”熱點”。 若能成功了解即辨識分析出其中的機制將會在生物學相關研究上有所貢獻。近期有許多的研究學著提出不同的方法來針對這個區塊”熱點”做預測分析。本篇研究提出了一個能針對在蛋白質交互作用接出面上有效的”熱點”做預測的系統，名為HotSpotFinder。此系統是經由擷取出蛋白質的模組結構、序列及空間上的資訊來進行交互作用接觸面”熱點”的預測分析，並透過兩步式資料萃取法來收斂特徵集，進而找到最佳化的特徵集。HotSpotFinder是由兩個預測分析器所組成，一個是由38個最佳特徵合成的特徵集所組成的預測器，稱作HotSpotFinder-Integrated；另一個是利用混合式系統概念所建立的預測器，稱作HotSpotFinder-Hybrid。與其他先前的研究相比，此系統除了在效能上有所提升外，即使在未包含於訓練資料集內的獨立資料集進行預測仍然維持著不錯的預測水準。;In a protein–protein interface, a small subset of residues contribute to the majority of the binding free energy, called the “hot spot”. Identifying and understanding hot spots and their mechanisms would have significant implications for bioinformatics and practical applications. Recently, many differences approaches have been used for predicted hot spot residues. We present an effective hot spot residues prediction system, HotSpotFinder, which contains motif recognition, sequential and spatial features and integrates feature set by two-step feature selection method. Through the two predictor of the system, called HotSpotFinder-Integrated and HotSpotFinder-Hybrid, to predict PPI hot spot residues. A total 38 optimal integrated feature and a novel system designed concept are provided and compared with other computational hot spot prediction models, HotSpotFinder offers significant performance improvement in terms of precision, MCC, F1 score and sensitivity, even in the independent dataset.
