CN111899792A - Method for screening small open reading frames with peptide coding capacity - Google Patents

Abstract

The invention belongs to the technical field of medicines, and particularly relates to a method for screening a small open reading frame with peptide coding capacity, which comprises the steps of screening prokaryotic genomes annotated by sORFs from a database by taking the GC content of the genomes as the basis, grouping the prokaryotic genomes according to the GC content of the genomes, screening out the sORFs with definite biological functions in each genome, performing redundancy removal treatment by using a CD-Hit program, sequentially taking the peptide coding sORFs in each genome as positive samples, taking the corresponding random scrambled sequences as negative samples, respectively predicting the sORFs in other genomes as a primary training set, and taking the genome with the best prediction effect of the GC content interval of each genome as a source genome of a final training set; and based on the screened training set, using the codon usage frequency of each sequence as a characteristic parameter, and using a classifier to train and complete screening. The method has good screening effect on prokaryotic organisms and eukaryotic organisms.

Description

Method for screening small open reading frames with peptide coding capacity

Technical Field

The invention belongs to the technical field of medicines, and particularly relates to a method for screening a small open reading frame with peptide coding capacity.

Background

In recent years, various omics sequencing technologies, such as genome and transcriptome, have been rapidly developed, and a great research model from qualitative to quantitative in life sciences has been promoted. However, since the last 20 years of the publication of human genome maps, people are increasingly aware that the complexity of genome interpretation is far beyond expectations, and more milestone genomic dark materials are discovered from protein-coding genes to non-coding RNAs, which pose serious challenges to researchers. In 2016, the topic group of Eric professor at texas university reported in the Science journal a small peptide DWORF encoded by a small open reading frame (srf) located on lncRNA, which is only 34 amino acids in length and plays an important role in myocardial contraction. The discovery of DWORF has led to unprecedented interest in sORFs and their encoded polypeptides (SEPs), a long-standing overlooked genomic element.

Small open reading frames (sORFs) with the capability of encoding polypeptides are important elements of genomes with important functions discovered in recent years, and a large number of researches prove that peptide encoding sORFs exist universally and play important biological functions at present, the sequence length of the peptide encoding sORFs is below 100 amino acids, but the research on the sORFs is in the primary stage due to low expression level and abundance, short sequence and lack of experimental technology, biological information algorithms and related database resources capable of effectively identifying the sORFs are also lacked, and an effective identification and screening method is also lacked. In order to avoid false positive of the screening result, the traditional gene screening method usually artificially sets 300 bases or 100 amino acids as the minimum gene length and has obvious species dependence. Therefore, the development of effective peptide coding sORFs prediction screening technology has important significance for the mining of functional polypeptides.

Disclosure of Invention

The research fully fuses sequence characteristics and machine learning technology, develops a method capable of efficiently predicting peptide coding sORFs, has a good screening effect on prokaryotic organisms and eukaryotic organisms, and can provide a new technology for functional peptide research in the future.

sORFs have short sequences, large research difficulty and lack of effective data sets, so that the screening technology of the sORFs is difficult and few methods are available at present. The invention provides a method for effectively screening peptide coding sORFs by establishing a targeted sORFs data set and combining a machine learning technology.

In order to solve the technical problems, the technical scheme adopted by the invention is as follows:

a method of screening for small open reading frames having peptide coding capacity, said method comprising the steps of:

screening prokaryotic genomes annotated by sORFs from a database by taking the GC content of the genomes as a basis, grouping the prokaryotic genomes according to the GC content of the genomes, screening out the sORFs with definite biological functions in each genome, performing redundancy removal by using a CD-Hit program, sequentially taking peptide coding sORFs in each genome as positive samples, taking corresponding random scrambled sequences as negative samples, respectively predicting the sORFs in the rest genomes as a primary training set, and taking the genome with the best prediction effect of the GC content interval of each genome as a source genome of a final training set; and based on the screened training set, using the codon use frequency of each sequence as a characteristic parameter, and using Logistic regression as a classifier to train and complete screening.

Further, the codon usage frequency cof in each sequence was calculated as follows:

wherein, c_iIndicates the number of each codon, and N indicates the length of the sequence.

Further, the grouping method according to the GC content of the genome comprises the following steps: the genome GC content is divided into 5 parts which are respectively less than 29 percent, 30 percent to 39 percent, 40 percent to 49 percent, 50 percent to 59 percent and more than or equal to 60 percent.

Further, the threshold for redundancy processing is 80%.

Further, the start and stop codons of the negative sample are identical to those of the positive sample.

Furthermore, five genomes with the best prediction effect in the GC content interval of each genome are screened, sORFs meeting the conditions in the five genomes and the scrambled sequences thereof are selected as a final training set, and sORFs in the rest genomes are selected as a test set.

Furthermore, the screening efficiency is evaluated by taking the accuracy and the Markov correlation coefficient as evaluation parameters.

Further, accuracy A_CCAnd the mahalanobis correlation coefficient MCC are respectively expressed as:

TP and FN represent the number of positive samples that are correctly predicted and mispredicted, respectively, and TN and FP represent the number of negative samples that are correctly predicted and mispredicted, respectively.

Further, the results were obtained by using the 64 values corresponding to the codon usage frequency as the input values of the logistic regression classifier.

Advantageous effects

The method for screening the peptide coding sORFs provided by the invention can quickly and accurately screen sORFs with the peptide coding, fully fuse genome data and a machine learning method, has high screening accuracy and provides a new thought for functional gene and polypeptide research.

The method has no species dependency, can be suitable for different species, and has the prediction accuracy and the Markov Correlation Coefficient (MCC) value of 98.36 percent and 0.9673 percent respectively for sORFs with 50-100 amino acids, and the prediction accuracy and the MCC value of 92.27 percent and 0.8454 percent respectively for sORFs with less than 50 amino acids. The method can provide effective research technology for screening and related research of peptide coding sORFs and coding functional polypeptides thereof.

Detailed Description

Example 1

The purpose is as follows: sORFs have short sequences, large research difficulty and lack of effective data sets, so that the screening technology of the sORFs is difficult and few methods are available at present. The invention provides a method for effectively screening peptide coding sORFs by establishing a targeted sORFs data set and combining a machine learning technology.

The technical scheme is as follows: in order to solve the technical problems, the technical scheme adopted by the invention is as follows:

1. a standard data set is established. The core of the method is to develop an effective data set screening strategy and establish a reliable peptide coding sORFs data set which can be universally used in different species. Therefore, the invention firstly screens and downloads 61 nuclear organism genomes with sORFs annotations from a Refseq database based on the GC content of the genomes, and divides the genomes into 5 parts according to the GC content of the genomes, wherein the 5 parts are < 29%, 30% -39%, 40% -49%, 50% -59% and > 60%. Then, sORFs with definite biological functions in each genome are screened out, and then redundancy removal processing (80% threshold value) is carried out by using a CD-Hit program to form a data set required by the work. On the basis, the peptide coding sORFs in each genome are sequentially used as positive samples, the corresponding random scrambling sequences are used as negative samples (the start codon and the stop codon of the negative samples are consistent with those of the positive samples), the positive samples are used as a primary training set to respectively predict the sORFs in the rest 60 genomes, and the genome with the best prediction effect of the GC content interval of each genome is used as a source genome of a final training set. Finally, the sORFs meeting the conditions in five genomes of NC _009089, NC _00313, NC _012962, NC _000913 and NC _008380 and the scrambled sequences thereof are used as a final training set, and the sORFs in the remaining 59 genomes are used as a test set.

2. And (4) a classifier. The present work uses Logistic regression as a coding/non-coding classifier.

3. And (4) characteristic parameters. The characteristic parameter of the algorithm is the percentage content of the codon.

The method specifically comprises the following steps:

1. a standard data set. Establishing a reliable and species-universal peptide-encoding sORFs dataset is the key point of the method. The invention takes the GC content of the genome as the basis, firstly screens and downloads 61 prokaryotic genomes with sORFs annotation from a Refseq database, and divides the prokaryotic genomes into 5 parts according to the GC content of the genomes, wherein the parts are < 29%, 30% -39%, 40% -49%, 50% -59% and more than or equal to 60%. Then, sORFs with definite biological functions in each genome were screened, and then redundancy removal processing (80% threshold) was performed by using the CD-Hit program to form a data set required for this work (Table 1). On the basis, the peptide coding sORFs in each genome are sequentially used as positive samples, the corresponding random scrambling sequences are used as negative samples (the start codon and the stop codon of the negative samples are consistent with those of the positive samples), the positive samples are used as a primary training set to respectively predict the sORFs in the rest 60 genomes, and the genome with the best prediction effect of the GC content interval of each genome is used as a source genome of a final training set. Finally, the sORFs meeting the conditions in five genomes of NC _009089, NC _00313, NC _012962, NC _000913 and NC _008380 and the scrambled sequences thereof are used as a final training set, and the sORFs in the remaining 59 genomes are used as a test set.

TABLE 1 information on peptides encoding sORFs satisfying the conditions in each genome

2. Characteristic parameter

For short sequences, most of the conventional characteristic parameters cannot be applied. In this work, we have tested codon usage frequency (cof), amino acid usage frequency (amf), and sequence complexity (sc) in each sequence, and each parameter is described as follows:

wherein, c_iDenotes the number of each codon, N denotes the sequence length, AA_jThe number of each amino acid is indicated.

3. Classifier

The method adopts Logistic regression (Logistic regression) as a classifier.

The classifier is a classification model for distinguishing positive and negative samples, after the training of the training set, a user inputs an sORF sequence, and the method can judge whether the input sequence has the peptide coding capacity according to the training of the classifier.

4. Evaluation of efficiency

In order to evaluate the screening efficiency of the method, the accuracy (accuracycacy, A) is respectively utilized_CC) And the Matthew's Correlation Coefficient (MCC) as evaluation parameters, which are respectively expressed as:

TP and FN represent the number of positive samples that are correctly predicted and mispredicted, respectively, and TN and FP represent the number of negative samples that are correctly predicted and mispredicted, respectively.

5. Efficiency of screening

In order to ensure the best screening effect, several characteristic parameters in step 2, namely three parameters of codon usage frequency (cof), amino acid usage frequency (amf) and sequence complexity (sc) in each sequence are respectively used as input characteristics for training, and then according to the results in table 2, the most efficient one is selected as the final characteristic parameter.

The screening efficiency of the present invention for different data sets is provided in table 2, and the results show that the screening effect for sORFs with a length of less than 50 amino acids is the best when codon usage preference is used, the screening effect for other sORFs is also more ideal, and the efficiency is the best when the sORFs are used as characteristic parameters, so that the codon usage frequency is finally used as the characteristic parameters in the method. The results were obtained using the 64 values between which the codon usage frequency corresponds as input values to a logistic regression classifier. In order to further show the stability of the screening efficiency of the method, table 3 provides the single prediction efficiency of the sORFs for each genome, the prediction accuracy and the MCC value mean values of the sORFs for each genome are 0.9777 and 0.9562 respectively, and the median values are 0.9846 and 0.9697 respectively, which shows that the method has good robustness.

TABLE 2 screening efficiency of peptide-encoded sORFs

TABLE 3 prediction efficiency of results from screening of peptide-encoded sORFs in different genomes

The method for screening the peptide coding sORFs gene provided by the invention sufficiently fuses genome data and a machine learning method, has high screening accuracy and provides a new thought for functional gene and polypeptide research.

Claims (9)

1. A method of screening for small open reading frames having peptide coding capacity, said method comprising the steps of:

screening prokaryotic genomes annotated by sORFs from a database by taking the GC content of the genomes as a basis, grouping the prokaryotic genomes according to the GC content of the genomes, screening out the sORFs with definite biological functions in each genome, performing redundancy removal by using a CD-Hit program, sequentially taking peptide coding sORFs in each genome as positive samples, taking corresponding random scrambled sequences as negative samples, respectively predicting the sORFs in the rest genomes as a primary training set, and taking the genome with the best prediction effect of the GC content interval of each genome as a source genome of a final training set; and based on the screened training set, using the codon use frequency of each sequence as a characteristic parameter, and using Logistic regression as a classifier to train and complete screening.

2. The method for screening for small open reading frames capable of encoding peptides according to claim 1, wherein the codon usage frequency cof in each sequence is calculated as follows:

wherein, c_iIndicates the number of each codon, and N indicates the length of the sequence.

3. The method for screening small open reading frames with peptide coding capacity according to claim 1, wherein the grouping according to the GC content of the genome is: the genome GC content is divided into 5 parts which are respectively less than 29 percent, 30 percent to 39 percent, 40 percent to 49 percent, 50 percent to 59 percent and more than or equal to 60 percent.

4. The method for screening for small open reading frames capable of encoding peptides of claim 1 wherein the threshold for redundant processing is 80%.

5. The method for selecting a small open reading frame with peptide coding ability according to claim 1, wherein the start and stop codons of the negative sample are identical to those of the positive sample.

6. The method of claim 1, wherein five genomes with the best prediction effect in the GC content interval of each genome are screened, the sORFs satisfying the conditions in the five genomes and the scrambled sequences thereof are selected as a final training set, and the sORFs in the rest genomes are selected as a test set.

7. The method for screening open reading frames with small peptide coding capacity according to claim 1, characterized in that the screening efficiency is evaluated by using accuracy and Markov correlation coefficient as evaluation parameters.

8. The method of claim 1 for screening for small open reading frames with peptide coding capacity with accuracy A_CCAnd the mahalanobis correlation coefficient MCC are respectively expressed as:

TP and FN represent the number of positive samples that are correctly predicted and mispredicted, respectively, and TN and FP represent the number of negative samples that are correctly predicted and mispredicted, respectively.

9. The method of claim 1, wherein the results are obtained by using 64 values corresponding to codon usage frequency as input values of a logistic regression classifier.