Method for predicting ctcf chromatin loop based on transcription factor sequence in ring extrusion model

Abstract

The present application relates to a method for predicting CTCF chromatin loop based on transcription factor sequence in loop extrusion model. It solves the problem of limited interpretability and inaccurate prediction of chromatin loop prediction method in the prior art. It comprises S1, constructing a data set; S2, key transcription factor sequence transformation; S3, establishing a CTCF chromatin loop classification model based on key transcription factor sequence; S4, model evaluation. The present application has the advantages that: only using key transcription factor sequence as input, the prediction of CTCF chromatin loop can be realized. The transcription factor binding sequence is innovatively encoded into the transcription factor sequence, and the pre-trained BERT model of natural language is used to decode the transcription factor sequence, so as to realize the prediction of CTCF chromatin loop, improve the prediction accuracy, and enhance the interpretability of the prediction model.

Description

Technical Field

[0001] This invention relates to the field of healthcare informatics technology, specifically to a method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model. Background Technology

[0002] Existing methods for predicting CTCF chromatin loops include machine learning methods (CTCF-MP, Loll ipop, CCIP) and mathematical modeling methods (LC models) based on sample genome sequences and epigenomic features. Existing algorithms rely on complex genome sequences and epigenomic features, including various DNA sequence compositions, histone modifications, gene expression RNA-seq signals, genome binding signals of the structural proteins CTCF and RAD21, loop length, and CTCF Motif orientation as input. They simply depend on the learning capabilities of machine learning or deep learning techniques, resulting in a black-box problem and limited interpretability. Methods based on simple mathematical models often require manual feature selection and construction, such as LC models, which may miss important information, leading to inaccurate predictions. Summary of the Invention

[0003] The purpose of this invention is to provide a method for predicting CTCF chromatin loops based on transcription factor sequences in a loop squash model, addressing the current limitations of chromatin loop prediction methods with poor interpretability of feature patterns.

[0004] To achieve the above objectives, the present invention employs the following technical solution: a method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model, characterized in that the method includes the following steps:

[0005] S1. Construct the dataset;

[0006] S2, key transcription factor sequence transformation;

[0007] S3. Establish a CTCF chromatin loop classification model based on key transcription factor sequences;

[0008] S4. Model Evaluation.

[0009] In step S1, chromatin loops are obtained using Hi-C experimental data, and the location coordinates of transcription factors in the genome are obtained using ChIP-seq data of transcription factors at the cell differentiation stage.

[0010] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S1 includes the following steps:

[0011] S11. Obtain all CTCF combination modes within a 1Mb range;

[0012] S12. Select chromatin loops verified by Hi-C experiments from all CTCF pairs as the positive set;

[0013] S13. Remove the data from the Positive Set and Bin Set garbage bins. What remains is the negative pool.

[0014] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S11 includes the following steps:

[0015] S111. First, all CTCFs in the SC-βorg stage of pancreatic cell differentiation are paired up within each chromosome.

[0016] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S12 includes the following steps:

[0017] S121. Remove all CTCF pairs outside the 1Mb range, retain the starting coordinates of the latter CTCF and the ending coordinates of the former CTCF, and call this dataset 1mb_ctcf_pairs. Then perform overlap analysis on all 1mb_ctcf_pairs with Hi-C data to find the regions that are both contained in 1mb_ctcf_pairs and have overlap in Hi-C data.

[0018] S122. Select the CTCF binding site with the highest peak value in the overlapping region and define it as the positive sample, i.e., the Positive Set.

[0019] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S13 includes the following steps:

[0020] S131. The remaining overlapping area data is put into the Bin Set database.

[0021] S132. Remove the positive set and bin data from all 1mb_ctcf_pairs, and the remaining data is the negative pool.

[0022] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S2 specifically includes the following steps:

[0023] S21. Download the CTCF motif from the JASPAR database: MA0139.1.transfac;

[0024] S22. Obtain directional data by combining FASTA files with CTCF genome sequences;

[0025] S23. Use the bedtools tool to integrate the narrowpeak files of the five transcription factors with the Hi-C loop data to generate a file containing transcription factor information and overlapping regions;

[0026] S24. The data was sorted and processed, and the arrangement of key transcription factors in the chromatin loop was finally constructed.

[0027] In step S3, when building the classification model, 80% of the samples are randomly selected as the training set and 20% as the independent test set. The model is then fine-tuned using the pre-trained BERT-Base-Uncased model for binary classification.

[0028] In the above-mentioned method for predicting CTCF chromatin loops based on the loop compression model, BERT is a natural language processing model, and BERT-Base-Uncased is a specific variant of the BERT model.

[0029] In the above-mentioned method for predicting CTCF chromatin loops based on the loop compression model, the model establishment includes the following steps:

[0030] S31. First, define a preprocessing function to mark the sequence and pad it with a fixed length;

[0031] S32. Next, define the training parameters and evaluation metrics.

[0032] Through the above steps, a classification model of CTCF chromatin loops based on BERT-Based-Uncased was established, and this classification model classifies key transcription factor sequences within the chromatin loops.

[0033] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S4 specifically includes the following steps:

[0034] S41. Obtain the predicted probabilities and true labels of the two models, LC model and BERT-Base-Uncased, and calculate the AUROC and AUPRC values.

[0035] S42. Logistic regression for TPR and FPR based on different thresholds;

[0036] S43. Plot the ROC curve using FPR as the x-axis and TPR as the y-axis.

[0037] S44. Measure the performance of the classifier and judge the accuracy of the classification results.

[0038] In step S41, the formulas for calculating the AUROC and AUPRC values ​​are as follows:

[0039]

[0040] In this context, TP, FP, TN, and FN represent the quantities of true positives, false positives, true negatives, and false negatives, respectively. TPR is the true positive rate, which is the proportion of positive samples that are correctly identified as positive out of all positive samples. FPR is the false positive rate, which is the proportion of negative samples that are incorrectly identified as positive out of all negative samples.

[0041] Compared with existing technologies, the advantages of this invention are:

[0042] 1. Using only key transcription factor sequences as input, it is possible to predict CTCF chromatin loops. This can be extended to other chromatin structure and transcription factor research areas, providing new tools and ideas for genomics and epigenomics research.

[0043] 2. By using innovative feature representation methods and fine-tuning with a pre-trained natural language model, the accuracy of CTCF chromatin ring prediction can be significantly improved.

[0044] 3. Innovatively, the binding sequence of transcription factors is encoded into a transcription factor sequence. Using a pre-trained BERT model based on natural language, the transcription factor sequence is decoded to predict CTCF chromatin loops, improving prediction accuracy while enhancing the interpretability of the prediction model. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the ring extrusion model in this invention;

[0046] Figure 2 This is a flowchart of the data processing in this invention;

[0047] Figure 3 This is a schematic diagram showing the test results of the Bert-Base-Uncased model parameters fine-tuned on the training set in this invention on an independent test set; Detailed Implementation

[0048] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0049] This method, based on the prediction of CTCF chromatin loops using transcription factor sequences in a loop compression model, includes the following steps:

[0050] S1. Construct the dataset;

[0051] S2, key transcription factor sequence transformation;

[0052] S3. Establish a CTCF chromatin loop classification model based on key transcription factor sequences;

[0053] S4. Model Evaluation.

[0054] In step S1, chromatin loops are obtained using Hi-C experimental data, and the location coordinates of transcription factors in the genome are obtained using ChIP-seq data of transcription factors in the SC-βorg stage of pancreatic cell differentiation.

[0055] The ring compression model is a widely accepted model for explaining the molecular mechanism of cohesin-mediated chromatin interactions. According to this model, specific protein complexes, such as CTCF and cohesin, bind to specific sites in chromatin and actively compress the chromatin to form a ring structure. Finally, WAPL is responsible for releasing cohesin from the chromatin, thereby ending the formation of the chromatin ring.

[0056] like Figure 1 As shown, according to the ring squeezing model, chromatin architecture proteins CTCF and cohesin bind to specific sites in the genome and actively squeeze chromatin to form a ring structure with the participation of ATP. Finally, the unloading protein WAPL is responsible for releasing cohesin from the chromatin, thus ending the chromatin ring squeezing.

[0057] Data processing flow as follows Figure 2 As shown:

[0058] a. Combine CTCF pairs to form a CTCF pairs dataset;

[0059] b. Obtain all CTCF combination modes within a 1Mb range;

[0060] c. Distribution of the positive and negative pools;

[0061] Sources of key transcription factor sequences: Hi-C loop and narrowpeak files of key transcription factors.

[0062] After the positive and negative sets (imbalanced datasets) sequences are input into the Bert-Base-Uncased model, AUROC and AUPRC are used as evaluation metrics.

[0063] The construction of the positive and negative sets in the dataset in step S1 includes the following steps:

[0064] S11. Obtain all CTCF combination modes within a 1Mb range;

[0065] S12. Select chromatin loops verified by Hi-C experiments from all CTCF pairs as the positive set;

[0066] S13. Remove the data from the Positive Set and Bin Set garbage bins. What remains is the negative pool.

[0067] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S11 includes the following steps:

[0068] S111. First, all CTCFs in the SC-βorg stage of pancreatic cell differentiation are paired up within each chromosome.

[0069] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S12 includes the following steps:

[0070] S121. Remove all CTCF pairs outside the 1Mb range, retain the starting coordinates of the latter CTCF and the ending coordinates of the former CTCF, and call this dataset 1mb_ctcf_pairs. Then perform overlap analysis on all 1mb_ctcf_pairs with Hi-C data to find the regions that are both contained in 1mb_ctcf_pairs and have overlap in Hi-C data.

[0071] S122. Select the CTCF binding site with the highest peak value in the overlapping region and define it as the positive sample, i.e., the Positive Set.

[0072] In the above method for predicting CTCF chromatin loops based on the loop compression model, step S13 includes the following steps:

[0073] S131. The remaining overlapping area data is put into the Bin Set database.

[0074] S132. Remove the Positive Set and Bin Set garbage data from all 1mb_ctcf_pairs. What remains is the Negative Set pool.

[0075] The sample sizes of the above datasets are shown in Table 1:

[0076]

[0077] Table 1 - Summary of Sample Size

[0078] As shown in Table 2, each transcription factor has a corresponding letter representation, where CTCF+ and CTCF- represent the forward and reverse directions of the CTCF binding site. This directionality plays an important role in the formation of chromatin loops, affecting the three-dimensional structure of the genome and the gene regulatory network.

[0079]

[0080] Table 2 - Alphabetical representation of key transcription factors

[0081] The steps to obtain the CTCF orientation using the Strom tool are as follows:

[0082] S21. Download the CTCF motif from the JASPAR database: MA0139.1.transfac;

[0083] S22. Obtain directional data by combining FASTA files with CTCF genome sequences;

[0084] S23. Use the bedtools tool to integrate the narrowpeak files of the five transcription factors with the Hi-C loop data to generate a file containing transcription factor information and overlapping regions;

[0085] S24. Sort and process the data to finally construct the arrangement of key transcription factors in the chromatin loop.

[0086] In step S3, when building the classification model, 80% of the samples are randomly selected as the training set and 20% as the independent test set. The model is then fine-tuned using the pre-trained BERT-Base-Uncased model for binary classification.

[0087] BERT stands for Natural Language Processing, and BERT-Base-Uncased is a specific variant of the BERT model.

[0088] BERT, or Bidirectional Encoder Representations from Transformers, is a natural language processing model developed by Google, designed to provide powerful pre-trained representations for a wide range of language understanding tasks.

[0089] In this experiment, the input to the model was the sequence of key transcription factors within the chromatin loop.

[0090] In detail, model building includes the following steps:

[0091] S31. First, define a preprocessing function to mark the sequence and pad it with a fixed length;

[0092] S32. Next, define the training parameters and evaluation metrics.

[0093] Through the above steps, a classification model of CTCF chromatin loops based on BERT-Based-Uncased was established, and this classification model classifies key transcription factor sequences within the chromatin loops.

[0094] Step S4 specifically includes the following steps:

[0095] S41. Obtain the predicted probabilities and true labels of the two models, LC model and BERT-Base-Uncased, and calculate the AUROC and AUPRC values.

[0096] S42. Logistic regression for TPR and FPR based on different thresholds;

[0097] S43. Plot the ROC curve using FPR as the x-axis and TPR as the y-axis.

[0098] S44. Measure the performance of the classifier and judge the accuracy of the classification results.

[0099] like Figure 3 As shown, the left graph is the ROC curve, with the horizontal axis representing the false positive rate and the vertical axis representing the true positive rate. The blue curve represents the LC model, and the orange curve represents the Bert-base-uncased model. A higher AUROC indicates that the Bert-base-uncased model outperforms the LC model in classification. The right graph is the PRC curve, with the horizontal axis representing recall and the vertical axis representing precision. A higher AUPRC also indicates that the Bert-base-uncased model performs better when dealing with imbalanced datasets.

[0100] The area under the ROC curve is called AUROC, which measures the classifier's performance and judges the accuracy of the classification results. Because the dataset in this experiment is imbalanced, with a positive-to-negative set ratio of 1:65 (positive samples far fewer than negative samples), the PRC curve is used to evaluate the model's performance. The horizontal axis of the PRC curve represents recall, and the vertical axis represents precision.

[0101] In step S41, the formulas for calculating the AUROC and AUPRC values ​​are as follows:

[0102]

[0103] In this context, TP, FP, TN, and FN represent the quantities of true positives, false positives, true negatives, and false negatives, respectively. TPR is the true positive rate, which is the proportion of positive samples that are correctly identified as positive out of all positive samples. FPR is the false positive rate, which is the proportion of negative samples that are incorrectly identified as positive out of all negative samples.

[0104] In summary, the principle of this embodiment is as follows: by collecting ChIP-seq data of five transcription factors in the SC-βorg stage of pancreatic cell differentiation, namely CTCF, NIBPL, WAPL, YY1, and RAD21, where CTCF is directional and denoted as CTCF+ and CTCF-, the distribution of different transcription factors in the genome is described. Each transcription factor is represented by a different letter, and transcription factor sequences are formed according to their binding sites. Hi-C experimental data are used to mark whether any two CTCF binding sites form chromatin loops, and a Bert pre-trained model is used to predict CTCF chromatin loops.

[0105] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model, characterized in that, This method includes the following steps: S1. Construct the dataset; S2, key transcription factor sequence transformation; S3. Establish a CTCF chromatin loop classification model based on key transcription factor sequences; S4. Model Evaluation; Step S1 includes the following steps: S11. Obtain all CTCF combination modes within a 1Mb range; S12. Select chromatin loops verified by Hi-C experiments from all CTCF pairs as the positive set; S13. Remove the data from the Positive Set and Bin Set garbage bins. What remains is the negative set pool. Step S11 includes the following steps: S111, First, all CTCFs in the SC-β org stage of pancreatic cell differentiation are paired up within each chromosome; Step S12 includes the following steps: S121. Remove all CTCF pairs outside the 1Mb range, retain the starting coordinates of the latter CTCF and the ending coordinates of the former CTCF, and call this dataset 1mb_ctcf_pairs. Then perform overlap analysis on all 1mb_ctcf_pairs with Hi-C data to find the regions that are both contained in 1mb_ctcf_pairs and have overlap in Hi-C data. S122. Select the CTCF binding site with the highest peak value in the overlapping region and define it as the positive sample, i.e., PositiveSet. Step S13 includes the following steps: S131. The remaining overlapping area data is put into the Bin Set database. S132. Remove the positive set and bin data from all 1mb_ctcf_pairs, and the remaining data is the negative pool.

2. The method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model according to claim 1, characterized in that, In step S1, chromatin loops are obtained using Hi-C experimental data, and the location coordinates of transcription factors in the genome are obtained using ChIP-seq data of transcription factors in the SC-βorg stage of pancreatic cell differentiation.

3. The method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model according to claim 1, characterized in that, Step S2 specifically includes the following steps: S21. Download the CTCF motif from the JASPAR database: MA0139.1.transfac; S22. Obtain directional data by combining FASTA files with CTCF genome sequences; S23. Use the bedtools tool to integrate the narrowpeak files of the five transcription factors with the Hi-C loop data to generate a file containing transcription factor information and overlapping regions; S24. The data was sorted and processed, and the arrangement of key transcription factors in the chromatin loop was finally constructed.

4. The method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model according to claim 1, characterized in that, In step S3, when building the classification model, 80% of the samples are randomly selected as the training set and 20% as the independent test set. The model is then fine-tuned using the pre-trained BERT-Base-Uncased model for binary classification.

5. The method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model according to claim 4, characterized in that, BERT is a natural language processing model, and BERT-Base-Uncased is a specific variant of the BERT model.

6. The method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model according to claim 5, characterized in that, Model building includes the following steps: S31. First, define a preprocessing function to mark the sequence and pad it with a fixed length; S32. Next, define the training parameters and evaluation metrics; Through the above steps, a classification model of CTCF chromatin loops based on BERT-Based-Uncased is established, and this classification model classifies key transcription factor sequences within the chromatin loops.

7. The method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41. Obtain the predicted probabilities and true labels of the two models, LC model and BERT-Base-Uncased, and calculate the AUROC and AUPRC values. S42. Logistic regression for TPR and FPR based on different thresholds; S43. Plot the ROC curve using FPR as the x-axis and TPR as the y-axis. S44. Measure the performance of the classifier and judge the accuracy of the classification results.

8. The method for predicting CTCF chromatin loops based on transcription factor sequences in a loop compression model according to claim 7, characterized in that, In step S41, the formulas for calculating the AUROC and AUPRC values ​​are as follows: ； ； ； In this context, TP, FP, TN, and FN represent the quantities of true positives, false positives, true negatives, and false negatives, respectively. TPR is the true positive rate, which is the proportion of positive samples that are correctly identified as positive out of all positive samples. FPR is the false positive rate, which is the proportion of negative samples that are incorrectly identified as positive out of all negative samples.

Images