【正文】 imilar to a substring of SD sequence, the algorithm calculates the alignment weight matrix of [3+l+2] bp size of window around the hit motif. ? To detect the context feature of start codon fragments around starts. ? Calculate the positional probability within the alignment windows around start codon with length of (4+3+15) bp. ? We may represent the weight matrix by wSD(k)(bi, i) for bi?{A, C, G, T}, where (k) means the kth iterative step and i means position within these alignment windows and (4+3+15)? i ?1. ? Despite the difficulty of unknown true start codons, we can reach an approximation through this weight matrix, because nucleotides occur more randomly around the false start codons. (3). Weight matrix for start codon context (4). Weights for potential start codons behind the leftmost start codon ? Not all the start codons have equal possibility to be selected as true gene start, different weights should be assigned to different start codons when they are investigated whether to be true translation initiation sites ? Note m is the index of start codons, define wm(k) as the weight of the mth start codon being true gene start site, k is the iterative step. ? Describes the likelihood for a start codon of order m counting from the left most one to be a true start site. ? For k=1, . in the first iterative step, as the initial condition, we set an equal weight to each wm(k) , . w1(1) = w2(1) =…=. (5). RBS score for start codon and the mostlikely start codon )(4kipP ?Lil ??? ?? ????321)(3 ,ljjkSD jbwP? ?? ????15341)(2 ,jjkS t a r t jbwP)(1kmwP ?ATG ATG P1 P2 P3 P4 STP …CCC TCGAAGC… ATG …AACAGGAGGATT… …AGGATT… ? ?4321l o g PPPPi ?????? Each of the above four measurements translates to a probability measure, then the bined score reads: ? Iteration: at each step with a set of given candidate TIS (. beginning with the leftmost start codon), check the scores {?i} (l≤ i≤ L) for all lmers occurred within the L bp upstream regions for each start codon, and select the maximum of {?i} as the RBS score for this start codon, i. e., ? ?iLilkmS ?? ??m a x)(? Compare the RBS score Sm(k) of different start codon and choose one with the highest score as the most likely candidate for the TIS. ? The kth iteration pletes when all candidate start sites are tested and updated. We then repeat the calculation of candidate motifs and hit motifs and all other probability measures with reference to the newly updated candidate TIS. The iteration begins at the next step. ? The iterations were repeated until the parameters were at least 99% identical to that of the previous iteration. Genome 16S rRNA Hit motifs No. 1 No. 2 No. 3 E. coli TA A GG A GG T GA AGGAG CAGGA GGAGA B. subtillis TA GA A A GGAGG GGAGG AAAGG AGGAG T. maritima GAAAGGAGGTG GAGGT ? ? H. influenzae TA A GG A GG T GA AAGGA ? ? M. jannaschii GG A GG TG AT C C AGGTG GGTGA ? ? The results suggest that the algorithm is rather effective to search the motifs associated with the SD sequences — almost each of the hit motifs is in good agreement with some substring of the reverse plement of the 3? end of 16S rRNA. (6). Convergence of selftrained model and the final parameters Table: Final hit motifs founded by MEDStart as potential 16S rRNA binding sites of various prokaryotes. Spacer distribution of the final hit motif with the highest ? for various prokaryotes. ‘AGGAG’ for E. coli ‘GGAGG’ for B. subtillis ‘GAGGT’ for T. maritima ‘AAGGA’ for H. influenzae ‘AGGTG’ for M. jannaschii MEDStart對翻譯調控信號特征的刻畫 MED探測到的枯草芽胞桿菌基因組中調控翻譯的多個信號“ GGAGG”、“ AAAGG”、“ AGGAG”以及它們的位置特異性。 RBSfinder (Salsberg et al., 2022) Postprocessor for GLIMMER GSFinder (Zhang Chunting et al., 2022) Postprocessor for ZCURVE MEDStart (She amp。 Zhu et al., 2022) Postprocessor for MED 張春霆 我國著名生物信息學家，天津大學中國科學院院士、第三世界科學院院士。 Steven Salzberg Senior Director of Bioinformatics, The Institute for Genomic Research, Johns Hopkins University, MEDStart的預測水平 MEDStart的預測水平 MEDStart的預測水平 MEDStart的預測水平 167。 原核基因的自動預測系統(tǒng) 1. EDP模型 ——刻畫 ORF序列整體編碼性與相似性 發(fā)展了對高 GC含量基因組的 EDP模型 2. TIS模型 ——刻畫基因上游區(qū)域的復雜序列特征 是基于 RBS模型的發(fā)展 定義基因翻譯起始的三種機制 刻畫基因翻譯起始信號的復雜性 考慮結構基因群的特征 考慮高 GC含量物種基因組的序列特征 3. 綜合運用 EDP模型、 TIS模型，發(fā)展了無監(jiān)督自學習的基因預測系統(tǒng) MED 流程圖 Naneq 古細菌 真核生物 細菌 MED模型參數(shù)揭示基因組轉錄、翻譯調控機制隨生物進化復雜程度的演化 翻譯調控信號 翻譯調控信號 翻譯調控信號 轉錄調控信號 轉錄調控信號 MED方法的特點 自由參數(shù)（～ 102個）少于傳統(tǒng)的 HMM方法，對學習集的依賴性小 HMM：～ 104個自由參數(shù)（如： GeneMark系統(tǒng)） 迭代自學習，大大少于其它方法的經(jīng)驗參數(shù)、預設參數(shù) 有利于新測序物種的基因組分析和注釋 預測精度達到并部分超過 GeneMark、 Glimmer等 模型參數(shù)具有非常明確的生物學意義，有利于基因組復雜結構信息的深刻理解 “事實上 ， 人類基因組計劃的巨大成功已經(jīng)表明 ，那些經(jīng)常用偏微分方程處理連續(xù)介質力學問題的傳統(tǒng)應用數(shù)學家對這一計劃所用到的數(shù)學方法并不熟悉 。 或許我們應該停下來思考一下 ，當我們將研究領域擴展到生命科學中去的時候 ，我們期望得到什么樣的結果 。 ” ——林家翹 《 應用數(shù)學的拓展 ——用一篇關于蛋白質分子的結構和 功能的動理論發(fā)展的論文來說明 》 （ 2022年第 2期“力學進展”） 感悟和體會