使用mRMR和随机森林的RNA结合蛋白序列预测法
93 浏览量
更新于2024-08-27
收藏 1.33MB PDF 举报
"本文介绍了使用随机森林与最小冗余最大相关性(mRMR)特征选择方法进行基于序列的RNA结合蛋白预测的研究论文。"
在生物信息学领域，预测RNA结合蛋白是一项极具挑战性的任务，因为这类蛋白质在细胞内起着至关重要的作用，如调控基因表达、参与信号传导等。尽管已有许多研究致力于解决这个问题，但预测的准确性仍然有待提高。这篇研究论文提出了一种新的方法，通过结合随机森林算法和mRMR特征选择技术，提升了预测RNA结合蛋白的精确度。
随机森林是一种集成学习方法，它构建了多个决策树并取其平均结果来提高预测性能。这种方法能处理大量特征，并且对过拟合有很好的抵抗能力。而mRMR（最小冗余最大相关性）是一种特征选择策略，它的目标是寻找一组最相关的特征，同时尽可能减少它们之间的相互冗余。这样的组合有助于提取最具有代表性和区分性的特征，从而提高模型的预测能力。
在本研究中，作者首先将氨基酸序列作为输入数据，然后应用mRMR方法来筛选出对预测最有价值的氨基酸属性。这些属性可能包括氨基酸的物理化学性质、序列位置的相对信息等。经过特征选择后，随机森林模型被训练用于分类，即判断一个蛋白质是否为RNA结合蛋白。通过这种方法，可以更有效地识别出具有RNA结合能力的蛋白质，从而帮助科学家理解其功能并进行后续实验验证。
该研究论文的重要性和贡献在于提供了一个准确的预测工具，这将有利于生物学研究，特别是在蛋白质功能注释、疾病相关性研究以及药物靶点发现等方面。此外，这种方法的通用性可能使其适用于其他生物大分子的预测问题，例如DNA结合蛋白或蛋白质-蛋白质相互作用的预测。
这篇论文提出的结合随机森林和mRMR特征选择的预测模型，为提高RNA结合蛋白预测的准确性开辟了新途径。通过深入挖掘氨基酸序列中的关键信息，该方法有望在生物信息学研究中发挥重要作用，并为生物医学研究提供有价值的预测工具。
BioMed Research International 3
correlation of RNA-binding residues in the query protein,
respectively. Furthermore, BP(2)formula represents the rel-
evance of the two RNA-binding residues combined with
dierent distances from 1 to −1and takes into account the
fact that the correlation value between two residues is smaller
when the distance is larger which proves the rationality of
the denition.
We also dened two nonbinding propensities for non-
binding proteins. e denitions of NBP(1)and NBP(2)are
similar to the denitions of BP(1)and BP(2). Consider
NBP
(
1
)
=
∑
𝑛
𝑖=1
RI
(
)
10
,
(3)
where and are the number of amino acids and the number
of nonbinding residues in this protein, respectively. RI() is
the reliability index of the prediction result of nonbinding
residue obtained from PRBR. Consider
NBP
(
2
)
=
∑
𝑁−1
𝑖=1
2
−𝑖+1
∑
𝑛(𝑖)
𝑘=1
RI
(
)
10
(
−1
)
,
(4)
where is the number of amino acids in this protein, ()
is the number of two nonbinding residues at a distance ,
and
RI() is the average value of the reliability index for
nonbinding residue and nonbinding residue +.
NBP(1) and NBP(2) describe the information of the
appearance and correlation of nonbinding residues in the
query protein, respectively, which are similar to BP(1) and
BP(2). We also used the reliability index because the pre-
diction result of nonbinding residues is applied in those
formulas.
2.2.2. Evolutionary Information Combined with Physicochem-
ical Properties (EIPP). Evolutionary information in the form
of a position-specic scoring matrix (PSSM) has been used
successfully to represent proteins in many applications, such
as prediction of DNA-binding residues [16–21] and RNA-
binding residues [15, 22, 23]. Here, PSSM proles were
generated using the PSI-BLAST program [24] to search the
nonredundant (NR) database through three iterations, with
0.001 as the -value cuto for multiple sequence alignment.
e PSSM scoring matrix has 20 ∗ elements, where is
the length of protein. However, dierent proteins may have
dierent numbers of amino acids. erefore, the PSSM could
not be used directly as feature in the prediction work because
all the machine learning methods require the input feature
to have a xed length. erefore, we generated a PSSM-400,
which has a vector of dimension of 400 from the PSSM.
PSSM-400 is composition of occurrences of each type of
amino acid corresponding to each type of amino acids in
sequences. We pooled all rows that belonged to the same
aminoacidinthisPSSMtoformanewmatrix.Wethen
converted each new matrix to a vector and added all the
normalized values in each column for the new matrixes.
erefore, we produced a 20-dimensional vector for each new
matrix to generate PSSM-400.
ephysicochemicalpropertyfeaturehasbeenused
eectively in many elds, such as the identication of
DNA\RNA-binding proteins [7, 9, 14, 25, 26] and the iden-
tication and prediction of protein-protein interactions [27].
us, an EIPP was generated by merging 20 amino acid
columns of the PSSM-400 into a single column containing
the information for a certain physicochemical property. Six
physicochemical properties that we used successfully in pre-
vious works [15] were considered for combining with PSSM-
400 to generate the EIPP: the pKa values of the amino group,
the pKa values of the carboxyl group [28], the molecular mass
[6], the lowest free energy [29], the Balaban index [30], and
theWienerindex[31].eentry
𝑎𝑘
of th type of amino acid
in a protein sequence for a certain physicochemical property
in EIPP was calculated with
𝑎𝑘
=
20
𝑖=1
𝑎
(
)
𝑘
(
)
,
(5)
where is the index of a certain physicochemical property,
is the index of the type of amino acids in the query protein
sequence, is the index of the type of na
¨
ıve amino acids,
𝑘
()
is the normalized value of the th type of na
¨
ıve amino acid
for the thtypeofaminoacidintheproteinsequenceof
the PSSM-400, and
𝑎
()is the normalized physicochemical
property values of for the th type amino acids. erefore,
thevectorsizeofEIPPfeatureis6×20.
2.2.3. Conjoint Triad (CT). Electrostatic and hydrophobic
interactions inuence protein-nucleic acid interactions and
may be reected by the dipoles and volumes of the side
chains of amino acids, respectively. Based on the dipoles
and volumes of the side chains, the 20 kinds of amino acids
could be clustered into seven classes [32]. Considering that
disulde bonds have no special eect on protein-nucleic acid
interactions, the unique amino acid cysteine in the seventh
class was put back to the third class in this study. erefore,
the 20 kinds of amino acids were clustered into six classes as
follows.Classa:Ala,Gly,andVal;Classb:Ile,Leu,Phe,and
Pro;Classc:Tyr,Met,r,Ser,andCys;Classd:His,Asn,
Gln, and Tpr; Class e: Arg and Lys; and Class f: Asp and Glu.
According to the similar feature construction method used
in [32], a protein is described by the conjoint triads feature
with 6×6×6=216dimensions, where each component
of the feature vector has the value of the frequency of the
corresponding triad.
As mentioned above, for each query protein, the vector
size of a feature is 4+120+216=340.
2.3. Algorithms to Classify and Measure a Classier’s Perfor-
mance. e random forest (RF) algorithm [12] is a classi-
cation algorithm that uses an ensemble of tree-structured
classiers, which has been used successfully in many applica-
tions for data classication and achieves high performance.
e random forest R package [33] was used to implement the
RF algorithm.
To evaluate the performance of the classier, a 5-fold
cross-validation procedure for the training dataset was used
in this research. During the procedure, we randomly divided
the data instances into ve parts. Four of these parts were
input into the RF to establish a model for classication,
剩余10页未读，继续阅读
2021-09-30 上传
2021-02-22 上传
2021-02-08 上传
2021-02-09 上传
2021-03-20 上传
2021-02-09 上传
2021-02-06 上传
2018-12-18 上传
2021-02-07 上传
weixin_38743481
- 粉丝: 696
- 资源: 4万+
我的内容管理
展开
最新资源
- 黑板风格计算机毕业答辩PPT模板下载
- CodeSandbox实现ListView快速创建指南
- Node.js脚本实现WXR文件到Postgres数据库帖子导入
- 清新简约创意三角毕业论文答辩PPT模板
- DISCORD-JS-CRUD：提升 Discord 机器人开发体验
- Node.js v4.3.2版本Linux ARM64平台运行时环境发布
- SQLight：C++11编写的轻量级MySQL客户端
- 计算机专业毕业论文答辩PPT模板
- Wireshark网络抓包工具的使用与数据包解析
- Wild Match Map: JavaScript中实现通配符映射与事件绑定
- 毕业答辩利器：蝶恋花毕业设计PPT模板
- Node.js深度解析：高性能Web服务器与实时应用构建
- 掌握深度图技术：游戏开发中的绚丽应用案例
- Dart语言的HTTP扩展包功能详解
- MoonMaker: 投资组合加固神器，助力$GME投资者登月
- 计算机毕业设计答辩PPT模板下载
