探索從頭設計IAPP結合性多肽之可行性

Abstract

蛋白質摺疊的過程在分子動力學中由於化學鍵震動的時間尺度與蛋白質之間結合的時間尺度差異巨大，因此要在有限時間內計算蛋白質尺度的抑制劑仍然不是一件容易的事，因此我們想要設計一套可以在個人電腦上計算並且有系統性的可以在適當的時間內快速搜尋蛋白質尺度抑制劑的方法。我們使用AlphaFold 3來建立蛋白質的初始模型，用來省去計算蛋白質複合物所需的龐大時間，這個建立模型的過程在網頁上可以在幾分鐘之內完成，用來完成初步的結構，並通過分子動力學計算AlphaFold 3所缺少的動力學數據，因為AlphaFold 3輸出結構是晶體結構，我們可以預期在適當的處理下，可以不需要過長的分子動力學流程，在將預測的蛋白質序列與動力學數據結合並且輸入適合的演算法中，用來製作下一次推演時AlphaFold 3模擬使用的序列。我們使用AlphaFold 3生成結構的優勢來補足分子動力學在結構探索上的缺點，再使用分子動力學的優勢來補足AlphaFold 3無法計算動力學數據的不足，我們主要處理的問題是因為多肽序列的每一個位置都有二十種可能的氨基酸，越長的序列會造成多肽的組合種類是一個天文數字，因此我們嘗試使用演算法將這個天文數字的組合中找到一個局部最佳解甚至是全域最佳解，將設計蛋白質等級的結合物從遙不可及變得可以嘗試。In the protein folding process in molecular dynamics, the time scale of chemical bond vibration is very different from the time scale of protein binding. Therefore, it is still not easy to calculate protein-scale inhibitors in a limited time. Therefore, we want to design a method that can be calculated on a personal computer and can quickly search for protein-scale inhibitors in a systematic manner within a reasonable time.We use AlphaFold 3 to build an initial model of the protein to save the huge amount of time required to calculate the protein complex. This model building process can be completed within a few minutes on the web page to complete the preliminary structure and calculate the dynamics data that AlphaFold 3 lacks through molecular dynamics. Because the output structure of AlphaFold 3 is a crystal structure, we can expect that with proper processing, there will be no need for an overly long molecular dynamics process. The predicted protein sequence will be combined with the dynamics data and input into a suitable algorithm to produce the sequence used by AlphaFold 3 simulation in the next deduction.We use the advantages of AlphaFold 3 in generating structures to make up for the shortcomings of molecular dynamics in structural exploration, and then use the advantages of molecular dynamics to make up for the inability of AlphaFold 3 to calculate dynamic data. The main problem we deal with is that there are twenty possible amino acids at each position in the peptide sequence. The longer the sequence, the more astronomical the number of peptide combinations. Therefore, we try to use algorithms to find a local optimal solution or even a global optimal solution among these astronomical combinations, making the design of protein-level binders from being out of reach to being possible.