A processing method and device for a multi-task processing model of mRNA

Abstract

The embodiment of the present invention relates to a processing method and device for a multi-task processing model for mRNA, the method comprising: constructing a multi-task processing model; collecting data on mRNA sequences of multiple species and corresponding Ribo-seq sequences, and training the multi-task processing model based on the collected data; and after the training, converting the mRNA sequence and Ribo-seq sequence input by the user into corresponding vectors and inputting them into the multi-task processing model for prediction to obtain the corresponding ribosome density / translation efficiency / translation extension rate / translation barrier point prediction vector; and correcting the Ribo-seq sequence input by the user according to the obtained prediction vector, and marking the ribosome high density area, efficient translation area, extension rate change area and potential barrier point of the mRNA sequence input by the user. The present invention can improve data acquisition efficiency, improve Ribo-seq data quality, and improve the generalization of predicted species.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a processing method and device for a multi-task processing model of mRNA. Background Art

[0002] Messenger Ribonucleic Acid (mRNA) plays a vital role in cells. Research on the associated information of mRNA, such as ribosome density, translation elongation rate, translation block point location, and ribosome profiling sequencing (Ribo-seq) data, will help reveal the dynamic changes of mRNA during translation and further understand the complex mechanism of gene expression regulation. However, currently, obtaining the ribosome density, translation elongation rate, and translation block point location information of a certain type of mRNA requires a large number of experiments and complex calculations, which makes data acquisition difficult and inefficient; and Ribo-seq data obtained through high-throughput sequencing cannot be directly used for data analysis due to the presence of technical artifacts. Summary of the invention

[0003] The purpose of the present invention is to provide a processing method, device, electronic device and computer-readable storage medium for a multi-task processing model of mRNA in view of the defects of the prior art. The present invention constructs a multi-task processing model for predicting the ribosome density, translation efficiency, translation extension rate and translation blocking point of an mRNA sequence; and performs big data collection on multiple types of mRNA sequences of multiple species and corresponding Ribo-seq sequences, and trains the multi-task processing model based on the collected data; and after the training, the mRNA sequence and Ribo-seq sequence input by the user are converted into corresponding vectors and input into the multi-task processing model for processing to obtain the corresponding ribosome density, translation efficiency, translation extension rate and translation blocking point prediction vectors, and the ribosome density prediction vectors are used to correct the Ribo-seq sequence input by the user, and the four types of prediction vectors obtained are used to mark the mRNA sequence input by the user for ribosome high density area, efficient translation area, extension rate change area and potential blocking point. Through the present invention, on the one hand, an end-to-end multi-task model can be directly used to predict the ribosome density, translation elongation rate and translation blocking points of mRNA, so as to shorten the data acquisition cycle and improve the data acquisition efficiency; on the other hand, the model prediction results can be directly used to correct the Ribo-seq data, so as to improve the quality of Ribo-seq data; on the other hand, the purpose of improving species generalization can be achieved through multi-species data training.

[0004] To achieve the above object, a first aspect of an embodiment of the present invention provides a processing method for a multi-task processing model of mRNA, the method comprising:

[0005] Construct a multi-task processing model for predicting ribosome density, translation efficiency, translation elongation rate and translation block points of mRNA sequences;

[0006] Data collection is performed on the mRNA sequences and corresponding Ribo-seq sequences of multiple species and a first data set is constructed based on the collected data; and model training is performed on the multi-task processing model based on the first data set; the mRNA sequence is formed by sequentially sorting a plurality of base units; the sequence length of the Ribo-seq sequence is consistent with the sequence length of the corresponding mRNA sequence; the Ribo-seq sequence is formed by sequentially sorting a plurality of ribosome footprint count values, and the ribosome footprint count values ​​correspond to the base units one by one;

[0007] After the model training is completed, the first mRNA sequence and the first Ribo-seq sequence input by the user are received;

[0008] Performing vector conversion on the first mRNA sequence and the first Ribo-seq sequence to obtain a corresponding first mRNA vector and a first Ribo-seq vector; and performing ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing on the first mRNA vector and the first Ribo-seq vector by the multi-task processing model to obtain a corresponding first ribosome density vector, a first translation efficiency vector, a first translation elongation rate vector and a first translation barrier point vector;

[0009] The first Ribo-seq sequence is corrected according to the first ribosome density vector; and the first mRNA sequence is marked with ribosome high density areas, efficient translation areas, extension rate change areas, and potential blocking points according to the first ribosome density vector, the first translation efficiency vector, the first translation elongation rate vector, and the first translation blocking point vector; and the corrected first Ribo-seq sequence and the marked first mRNA sequence are fed back to the user.

[0010] Preferably, the multi-task processing model is used to perform ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing according to the input mRNA vector A and Ribo-seq vector B and output the corresponding ribosome density prediction vector PD, translation efficiency prediction vector PE, translation elongation rate prediction vector PV and translation barrier point prediction vector PS;

[0011] The mRNA vector A corresponds to an mRNA sequence, and is encoded by multiple bases.i Sort by order, 1≤i≤N A , N A is the sequence length of the current mRNA sequence; the base encoding a i One-to-one correspondence with the base units of the current mRNA sequence;

[0012] The Ribo-seq vector B corresponds to the Ribo-seq sequence of the current mRNA sequence, and is represented by N A Footprint code b i The footprint code b i One-to-one correspondence with the ribosome footprint count value of the current Ribo-seq sequence and with the base encoding a i One to one correspondence;

[0013] The ribosome density prediction vector PD is composed of N A The ribosome density d i The ribosome density d i With the base coding a i One to one correspondence;

[0014] The translation efficiency prediction vector PE is composed of N A Translation efficiency e i The translation efficiency e i With the base coding a i One to one correspondence;

[0015] The translation elongation rate prediction vector PV is composed of N A The translation elongation rate v i The translation elongation rate v i With the base coding a i One to one correspondence;

[0016] The translation obstacle point prediction vector PS is composed of N A Translation point types i The translation point type s i With the base coding a i One-to-one correspondence; the translation point type s i There are two types: obstruction points and non-obstruction points;

[0017] The first model input end and the second model input end of the multi-task processing model are respectively used to receive the corresponding mRNA vector A and the Ribo-seq vector B, and the first model output end, the second model output end, the third model output end and the fourth model output end are respectively used to output the corresponding ribosome density prediction vector PD, the translation efficiency prediction vector PE, the translation elongation rate prediction vector PV and the translation blocking point prediction vector PS;

[0018] The multi-task processing model includes a feature fusion module, a Uni-RNA model, a ribosome density prediction head, a translation efficiency prediction head, a translation elongation rate prediction head and a translation blocking point prediction head;

[0019] The first and second input ends of the feature fusion module are respectively connected to the corresponding first and second model input ends, and the output end is connected to the input end of the Uni-RNA model; the output end of the Uni-RNA model is respectively connected to the input ends of the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation barrier point prediction head; the output ends of the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation barrier point prediction head are respectively connected to the corresponding first, second, third and fourth model output ends;

[0020] The feature fusion module is used to encode the mRNA vector A and the Ribo-seq vector B respectively according to the encoding method of the input embedding layer corresponding to the Uni-RNA model to generate the corresponding first and second encoding vectors; and send the Hadamard product vector of the first and second encoding vectors to the Uni-RNA model as the corresponding fusion feature vector X; the fusion feature vector X is composed of N A Fusion features x i Arranged in order;

[0021] The Uni-RNA model has completed pre-training; the Uni-RNA model is used to perform feature encoding processing on the fusion feature vector X to obtain a corresponding encoding feature vector Y and send it to the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation blocking point prediction head respectively;

[0022] The ribosome density prediction head is implemented based on a type of nonlinear regression prediction model; the ribosome density prediction head is used to predict the ribosome density at each base unit position according to the encoding feature vector Y to obtain the corresponding ribosome density prediction vector PD; the nonlinear regression prediction model at least includes an MLP model, an NNR model, a GBDT model and an XGBoost model;

[0023] The translation efficiency prediction head is implemented based on a type of nonlinear regression prediction model; the translation efficiency prediction head is used to predict the translation efficiency at each base unit position according to the encoding feature vector Y to obtain the corresponding translation efficiency prediction vector PE;

[0024] The translation elongation rate prediction head is implemented based on a type of nonlinear regression prediction model; the translation elongation rate prediction head is used to predict the translation elongation rate at each base unit position according to the encoding feature vector Y to obtain the corresponding translation elongation rate prediction vector PV;

[0025] The translation barrier point prediction head is implemented based on a binary classification model; the translation barrier point prediction head is used to perform binary classification identification on whether each base unit position point is a translation barrier point according to the encoding feature vector Y to obtain the corresponding translation barrier point prediction vector PS; the binary classification model at least includes a binary classifier implemented based on an MLP model, a binary classifier implemented based on a CNN model, a binary classifier implemented based on a decision tree model, and a binary classifier implemented based on a random forest model.

[0026] Preferably, the first data set covers multiple types of mRNA molecules of multiple species, and is composed of multiple first data records; each of the first data records corresponds to a type of mRNA molecules of a type of species;

[0027] The first data record includes a first mRNA tag sequence, a first Ribo-seq tag sequence, a first ribosome density tag vector, a first translation efficiency tag vector, a first translation elongation rate tag vector, and a first translation blocking point tag vector;

[0028] The first mRNA tag sequence is composed of multiple first tag base codes; the first Ribo-seq tag sequence is composed of multiple first tag footprint codes; the first ribosome density tag vector is composed of multiple first tag ribosome densities; the first translation efficiency tag vector is composed of multiple first tag translation efficiencies; the first translation elongation rate tag vector is composed of multiple first tag translation elongation rates; the first translation blocking point tag vector is composed of multiple first tag translation point types; the first tag translation point type includes two types: blocking points and non-blocking points; the first tag footprint code, the first tag ribosome density, the first tag translation efficiency, the first tag translation elongation rate and the first tag translation point type all correspond to the first tag base code one by one.

[0029] Preferably, the performing model training on the multi-task processing model based on the first data set specifically includes:

[0030] Step 41, based on a preset first segmentation ratio, randomly split the first data set into two sub-data sets recorded as a corresponding first training set and a first evaluation set;

[0031] Wherein, both the first training set and the first evaluation set are composed of a plurality of the first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio;

[0032] Step 42, extracting the first first data record of the first training set as the corresponding current training record;

[0033] Step 43, performing vector conversion on the first mRNA label sequence and the first Ribo-seq label sequence of the current training record to obtain the corresponding first training mRNA vector and the first training Ribo-seq vector; and recording the first ribosome density label vector, the first translation efficiency label vector, the first translation elongation rate label vector and the first translation blocking point label vector of the current training record as the corresponding first ribosome density label vector PD tag , the first translation efficiency label vector PE tag , the first translation elongation rate label vector PV tag and the first translation obstruction point label vector PS tag ;

[0034] Step 44, calculate the mean of all the first label footprint codes of the first training Ribo-seq vector to obtain the corresponding current mean code; and use the largest and smallest first label footprint codes in the first training Ribo-seq vector as the corresponding current maximum code and current minimum code; and record the coding range from the current minimum code to the current mean code, and the coding range from the current mean code to the current maximum code as the corresponding low-order coding range and high-order coding range; and randomly select an integer between the preset positive and negative integer range [-R1, R1] as the corresponding current A random number R2 is selected; and when the random number R2 is greater than 0, R2 first label footprint codes are randomly selected from the first training Ribo-seq vector as corresponding random scrambled code objects; and each random scrambled code object is compared with the current average code, if the current random scrambled code object is less than or equal to the current average code, a code is randomly selected from the high-order code range to modify the current random scrambled code object, and if the current random scrambled code object is greater than the current average code, a code is randomly selected from the low-order code range to modify the current random scrambled code object;

[0035] Step 45, inputting the first training mRNA vector and the first training Ribo-seq vector into the multi-task processing model to perform ribosome density, translation efficiency, translation elongation rate and translation blocking point prediction processing to obtain the corresponding first ribosome density prediction vector PD pre , the first translation efficiency prediction vector PE pre , the first translation elongation rate prediction vector PV pre and the first translation obstacle prediction vector PS pre ;

[0036] Step 46: transform the first ribosome density prediction vector PD pre and the first ribosome density label vector PD tag Substitute the preset first loss function L D , and the first translation efficiency prediction vector PE pre and the first translation efficiency label vector PE tag Substitute the preset second loss function L E , and the first translation elongation rate prediction vector PV pre and the first translation elongation rate tag vector PV tag Bring in the preset third loss function L V , and the first translation barrier point prediction vector PS pre and the first translation obstruction point label vector PS tag Substitute the preset fourth loss function L S , and by the first loss function L D 、The second loss function L E , the third loss function L V And the fourth loss function L S Add together to form the corresponding first model loss function L M ;

[0037] Among them, the first model loss function L M for:

[0038] L M =L D (PD pre ,PD tag )+L E (PE pre ,PE tag )+L V (PV pre ,PV tag )+L S (PS pre ,PS tag );

[0039] The first loss function L D 、The second loss function L E And the third loss function L V At least including MSE loss function, MAE loss function, Huber loss function, quantile loss function and SmoothL1 loss function; the fourth loss function L S At least including cross entropy loss function, logarithmic loss function and hinge loss function;

[0040] Step 47, based on a preset first model optimizer and a preset first iteration threshold, moves toward making the first loss function L D 、The second loss function L E , the third loss function L V , the fourth loss function L S And the first model loss function L M Performing a round of iterative modulation on the model parameters of the multi-task processing model in a direction in which the minimum value can be achieved;

[0041] Wherein, the first model optimizer includes at least an SGD optimizer and an ADAM optimizer;

[0042] Step 48, identifying whether the current training record is the last first data record of the first training set; if so, going to step 49; if not, extracting the next first data record of the first training set as the new current training record and returning to step 43;

[0043] Step 49, perform a round of traversal on all the first data records of the first evaluation set; and in this round of traversal, use the first data record currently traversed as the corresponding current evaluation record; and perform vector conversion on the first mRNA tag sequence and the first Ribo-seq tag sequence of the current evaluation record to obtain the corresponding second training mRNA vector and the second training Ribo-seq vector; and input the second training mRNA vector and the second training Ribo-seq vector into the multi-task processing model for ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing to obtain the corresponding second ribosome density prediction vector, second translation efficiency prediction vector, second translation elongation rate prediction vector and second translation barrier point prediction vector; and the second ribosome density prediction vector, the second translation efficiency prediction vector and the second translation elongation rate prediction vector are composed of The corresponding first prediction tensor is composed of the second translation barrier point prediction vector as the corresponding second prediction tensor; and the first ribosome density label vector, the first translation efficiency label vector and the first translation elongation rate label vector of the current training record form a corresponding first label tensor, and the first translation barrier point label vector of the current training record is used as the corresponding second label tensor; and the first prediction tensor and the first label tensor form a corresponding first prediction-label pair, and the second prediction tensor and the second label tensor form a corresponding second prediction-label pair; and at the end of this round of traversal, all the first prediction-label pairs obtained are brought into the preset first evaluation function for calculation to obtain the corresponding first evaluation value; and the precision, recall and F1 score evaluations are performed on all the second prediction-label pairs obtained to obtain the corresponding first precision, first recall and first F1 score;

[0044] Wherein, the first evaluation function includes at least an MSE function and an RMSE function;

[0045] Step 50, identifying the first evaluation value, the first precision, the first recall rate and the first F1 score; if the first evaluation value, the first precision, the first recall rate or the first F1 score does not satisfy the corresponding evaluation value range, precision range, recall rate range or F1 score range, returning to step 41 to continue training; if the first evaluation value, the first precision, the first recall rate and the first F1 score all satisfy their respective corresponding evaluation value range, precision range, recall rate range and F1 score range, stopping training and confirming that the model training of the multi-task processing model is completed.

[0046] Preferably, the correcting the first Ribo-seq sequence according to the first ribosome density vector specifically includes:

[0047] Step 51, normalizing all ribosome densities of the first ribosome density vector to obtain a corresponding first normalized sequence; and normalizing all ribosome footprint count values ​​of the first Ribo-seq sequence to obtain a corresponding second normalized sequence;

[0048] Wherein, the first and second normalized sequences are each composed of a plurality of corresponding first and second normalized values; the first and second normalized values ​​correspond to each other one by one;

[0049] Step 52, performing absolute difference calculation on the first and second normalized values ​​corresponding to each group to obtain a corresponding first absolute difference; and comparing each of the first absolute differences with a preset absolute difference threshold, if the first absolute difference currently compared exceeds the absolute difference threshold, recording the ribosome footprint count value corresponding to the current first absolute difference in the first Ribo-seq sequence as a corresponding count value to be corrected;

[0050] Step 53, identifying whether the total number of count values ​​to be corrected is 0;

[0051] Step 54: if the total number of the count values ​​to be corrected is 0, the first Ribo-seq sequence without modification is directly output as the corrected first Ribo-seq sequence;

[0052] Step 55, if the total number of the count values ​​to be corrected is greater than 0, then the corresponding corrected count values ​​are obtained by performing a denormalization calculation based on the first normalized values ​​corresponding to each of the count values ​​to be corrected; and the corresponding count values ​​to be corrected in the first Ribo-seq sequence are reset based on each of the corrected count values; and each reset ribosomal footprint count value is marked with a correction label on the first Ribo-seq sequence; and the modified and marked first Ribo-seq sequence is output as the corrected first Ribo-seq sequence.

[0053] Preferably, the step of marking the first mRNA sequence with a high ribosome density region, a high efficiency translation region, a region with a variable elongation rate, and a potential blocking point according to the first ribosome density vector, the first translation efficiency vector, the first translation elongation rate vector, and the first translation blocking point vector specifically includes:

[0054] Step 61, recording all ribosome densities exceeding a preset first density threshold in the first ribosome density vector as corresponding first high densities; and marking all base unit fragments corresponding to two or more consecutive first high densities in the first mRNA sequence as corresponding ribosome high density regions;

[0055] Step 62, recording the translation efficiencies exceeding the preset first efficiency threshold in the first translation efficiency vector as corresponding first high efficiencies; and marking the base unit fragments corresponding to two or more consecutive first high efficiencies in the first mRNA sequence as corresponding high-efficiency translation regions;

[0056] Step 63, performing differential calculation on two adjacent translation elongation rates in the first translation elongation rate vector to obtain corresponding first differential rates to form a corresponding first differential sequence; and recording the first differential rates in the first differential sequence that exceed a preset reasonable differential rate range as corresponding abnormal differential rates; and marking the base unit fragments consisting of two base units corresponding to each of the abnormal differential rates in the first mRNA sequence as a corresponding extension rate change region;

[0057] Step 64, record all translation point types that are blocking points in the first translation blocking point vector as corresponding blocking point types; and mark all base units corresponding to each of the blocking point types in the first mRNA sequence as a corresponding potential blocking point.

[0058] A second aspect of an embodiment of the present invention provides a device for implementing the processing method of the multi-task processing model for mRNA described in the first aspect, the device comprising: a model construction module, a model training module, a data receiving module, a model prediction module and an analysis output module;

[0059] The model building module is used to build a multi-task processing model for predicting ribosome density, translation efficiency, translation elongation rate and translation blocking points of mRNA sequences;

[0060] The model training module is used to collect data on the mRNA sequences and corresponding Ribo-seq sequences of multiple species and construct a first data set based on the collected data; and to perform model training on the multi-task processing model based on the first data set; the mRNA sequence is formed by sequentially sorting a plurality of base units; the sequence length of the Ribo-seq sequence is consistent with the sequence length of the corresponding mRNA sequence; the Ribo-seq sequence is formed by sequentially sorting a plurality of ribosome footprint count values, and the ribosome footprint count values ​​correspond to the base units one by one;

[0061] The data receiving module is used to receive a first mRNA sequence and a first Ribo-seq sequence input by a user after the model training is completed;

[0062] The model prediction module is used to perform vector conversion on the first mRNA sequence and the first Ribo-seq sequence to obtain the corresponding first mRNA vector and the first Ribo-seq vector; and the multi-task processing model performs ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing on the first mRNA vector and the first Ribo-seq vector to obtain the corresponding first ribosome density vector, first translation efficiency vector, first translation elongation rate vector and first translation barrier point vector;

[0063] The analysis output module is used to correct the first Ribo-seq sequence according to the first ribosome density vector; and mark the first mRNA sequence with high ribosome density areas, high-efficiency translation areas, extension rate variation areas and potential blocking points according to the first ribosome density vector, the first translation efficiency vector, the first translation extension rate vector and the first translation blocking point vector; and feed back the corrected first Ribo-seq sequence and the marked first mRNA sequence to the user.

[0064] A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0065] The processor is used to be coupled to the memory, read and execute instructions in the memory, so as to implement the method steps described in the first aspect above;

[0066] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0067] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions. When the computer instructions are executed by a computer, the computer executes the instructions of the method described in the first aspect above.

[0068] The embodiment of the present invention provides a processing method, device, electronic device and computer-readable storage medium for a multi-task processing model for mRNA. As can be seen from the above content, the embodiment of the present invention constructs a multi-task processing model for predicting the ribosome density, translation efficiency, translation extension rate and translation blocking point of an mRNA sequence; and performs big data collection on multiple types of mRNA sequences of multiple species and corresponding Ribo-seq sequences, and trains the multi-task processing model based on the collected data; and after the training, the mRNA sequence and Ribo-seq sequence input by the user are converted into corresponding vectors and input into the multi-task processing model for processing to obtain the corresponding ribosome density, translation efficiency, translation extension rate and translation blocking point prediction vectors, and use the ribosome density prediction vectors to correct the Ribo-seq sequence input by the user, and use the obtained four types of prediction vectors to mark the mRNA sequence input by the user for ribosome high density area, efficient translation area, extension rate change area and potential blocking point. Through the implementation of the present invention, on the one hand, an end-to-end multi-task model can be directly used to predict the ribosome density, translation elongation rate and translation blocking points of mRNA, thereby shortening the data acquisition cycle and improving the data acquisition efficiency; on the other hand, the model prediction results can be directly used to correct the Ribo-seq data and analyze and mark the mRNA sequence, which not only improves the quality of Ribo-seq data but also adds convenient data analysis methods; on the other hand, the species generalization of the model is also improved through multi-species data training. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A schematic diagram of a processing method for a multi-task processing model of mRNA provided in Example 1 of the present invention;

[0070] Figure 2 A module structure diagram of a multi-task processing model provided in the first embodiment of the present invention;

[0071] Figure 3 A module structure diagram of a processing device for a multi-task processing model of mRNA provided in Embodiment 2 of the present invention;

[0072] Figure 4 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0074] Embodiment 1 of the present invention provides a processing method for a multi-task processing model of mRNA, such as Figure 1 As shown in the schematic diagram of a processing method for a multi-task processing model of mRNA provided in the first embodiment of the present invention, the method mainly comprises the following steps:

[0075] Step 1: Build a multi-task processing model for predicting ribosome density, translation efficiency, translation elongation rate and translation blocking points of mRNA sequences.

[0076] Here, the mRNA sequence of the embodiment of the present invention is a one-dimensional molecular sequence of a class of mRNA molecules of a class of species, which is composed of a plurality of base units arranged in sequence. The sequence length of the Ribo-seq sequence corresponding to each mRNA sequence in the embodiment of the present invention should be consistent with the sequence length of the corresponding mRNA sequence; and each Ribo-seq sequence should be composed of a plurality of ribosomal footprint (RF) count values ​​(also known as ribosome coverage) arranged in sequence, and the ribosomal footprint count values ​​in the Ribo-seq sequence should correspond one-to-one to the base units in the corresponding mRNA sequence.

[0077] The multi-task processing model of the embodiment of the present invention is used to perform ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing according to the input mRNA vector A and Ribo-seq vector B, and output the corresponding ribosome density prediction vector PD, translation efficiency prediction vector PE, translation elongation rate prediction vector PV and translation barrier point prediction vector PS.

[0078] The above mRNA vector A corresponds to an mRNA sequence, which is encoded by multiple bases. i Sort by order, 1≤i≤N A , N A is the sequence length of the current mRNA sequence; base code a i Here, the embodiment of the present invention uses a one-hot encoding method to encode each base unit to obtain the corresponding base code a i .

[0079] The above Ribo-seq vector B corresponds to the Ribo-seq sequence of the current mRNA sequence, which is represented by NA Footprint code b i Sort by order; footprint code b i One-to-one correspondence with the ribosome footprint count value of the current Ribo-seq sequence and the base encoding a i Here, the embodiment of the present invention adopts the value copy method by default to directly use each ribosome footprint count value as the corresponding footprint code b i .

[0080] The above ribosome density prediction vector PD is composed of N A The ribosome density d i Sort by order; ribosome density d i With base code a i One to one correspondence.

[0081] The above translation efficiency prediction vector PE is composed of N A Translation efficiency e i Sort by order; translation efficiency e i With base code a i One to one correspondence.

[0082] The above translation elongation rate prediction vector PV is composed of N A The translation elongation rate v i Sequential sorting; translation elongation rate v i With base code a i One to one correspondence.

[0083] The above translation obstacle point prediction vector PS is composed of N A Translation point types i Sort by order; translation point type s i With base code a i One-to-one correspondence; translation point type s i There are two types: obstruction points and non-obstruction points.

[0084] like Figure 2 As shown in the module structure diagram of the multi-task processing model provided in the first embodiment of the present invention, the first model input end and the second model input end of the multi-task processing model of the embodiment of the present invention are respectively used to receive the corresponding mRNA vector A and Ribo-seq vector B, and the first model output end, the second model output end, the third model output end and the fourth model output end are respectively used to output the corresponding ribosome density prediction vector PD, the translation efficiency prediction vector PE, the translation elongation rate prediction vector PV and the translation barrier point prediction vector PS.

[0085] The model components of this multi-task processing model include: feature fusion module, Uni-RNA model, ribosome density prediction head, translation efficiency prediction head, translation elongation rate prediction head and translation blocking point prediction head.

[0086] The connection relationship between each model component is as follows: the first and second input ends of the feature fusion module are respectively connected to the corresponding first and second model input ends, and the output end is connected to the input end of the Uni-RNA model; the output end of the Uni-RNA model is respectively connected to the input ends of the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation barrier point prediction head; the output ends of the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation barrier point prediction head are respectively connected to the corresponding first, second, third and fourth model output ends.

[0087] The component functions of each model component are as follows.

[0088] 1) Feature fusion module:

[0089] The feature fusion module of the embodiment of the present invention is used to encode the mRNA vector A and the Ribo-seq vector B respectively according to the encoding method of the input embedding layer corresponding to the Uni-RNA model to generate the corresponding first and second encoding vectors; and send the Hadamard product vector of the first and second encoding vectors to the Uni-RNA model as the corresponding fused feature vector X; the fused feature vector X here is composed of N A Fusion features x i Sorted in order.

[0090] 2) Uni-RNA model:

[0091] The Uni-RNA model of the embodiment of the present invention is used to perform feature encoding processing on the fused feature vector X to obtain the corresponding encoded feature vector Y and send it to the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation barrier prediction head respectively.

[0092] It should be noted that the Uni-RNA model of the embodiment of the present invention is a feature encoding model implemented based on the transformer encoder structure, and the Uni-RNA model has completed pre-training; the model structure, function and pre-training scheme of this model have been disclosed in the technical document "UNI-RNA: UNIVERSAL PRE-TRAINED MODELS REVOLUTIONIZE RNARESEARCH". It can be seen from the technical document that the Uni-RNA model can be used to perform base-level deep learning on an RNA sequence and output a corresponding encoding feature, namely, the encoding feature vector Y.

[0093] 3) Ribosome density prediction head:

[0094] The ribosome density prediction head of the embodiment of the present invention is implemented based on a type of nonlinear regression prediction model; the ribosome density prediction head is used to predict the ribosome density at each base unit position according to the encoding feature vector Y to obtain the corresponding ribosome density prediction vector PD.

[0095] Here, the nonlinear regression prediction model mentioned in the embodiment of the present invention includes at least an MLP model, an NNR model, a GBDT model and an XGBoost model.

[0096] 4) Translation efficiency prediction head:

[0097] The translation efficiency prediction head of the embodiment of the present invention is implemented based on another type of nonlinear regression prediction model; the translation efficiency prediction head is used to predict the translation efficiency at each base unit position according to the encoding feature vector Y to obtain the corresponding translation efficiency prediction vector PE.

[0098] 5) Translation elongation rate prediction head:

[0099] The translation elongation rate prediction head of the embodiment of the present invention is implemented based on another type of nonlinear regression prediction model; the translation elongation rate prediction head is used to predict the translation elongation rate at each base unit position according to the encoding feature vector Y to obtain the corresponding translation elongation rate prediction vector PV.

[0100] 6) Translation blocking point prediction head:

[0101] The translation barrier point prediction head of the embodiment of the present invention is implemented based on a binary classification model; the translation barrier point prediction head is used to perform binary classification identification on whether each base unit position point is a translation barrier point according to the encoding feature vector Y to obtain the corresponding translation barrier point prediction vector PS.

[0102] Here, the binary classification model mentioned in the embodiment of the present invention includes at least a binary classifier implemented based on an MLP model, a binary classifier implemented based on a CNN model, a binary classifier implemented based on a decision tree model, and a binary classifier implemented based on a random forest model.

[0103] Step 2, collecting data on mRNA sequences and corresponding Ribo-seq sequences of multiple species and constructing a first data set based on the collected data; and training a multi-task processing model based on the first data set;

[0104] Specifically comprising: step 21, collecting data on mRNA sequences and corresponding Ribo-seq sequences of multiple species and constructing a first data set based on the collected data;

[0105] Here, the embodiment of the present invention constructs a cross-species standard data set, i.e., the first data set, by performing big data collection on multiple types of mRNA+Ribo-seq+gene expression data (such as ribosome density, translation efficiency, translation elongation rate, and translation blocking points) of multiple species; the first data set of the embodiment of the present invention covers multiple types of mRNA molecules of multiple species and is composed of multiple first data records;

[0106] Each first data record in the data set corresponds to a class of mRNA molecules of a class of species, and the record data items consist of a first mRNA tag sequence, a first Ribo-seq tag sequence, a first ribosome density tag vector, a first translation efficiency tag vector, a first translation elongation rate tag vector and a first translation blocking point tag vector; wherein the first mRNA tag sequence consists of a plurality of first tag base codes; the first Ribo-seq tag sequence consists of a plurality of first tag footprint codes; the first ribosome density tag vector consists of a plurality of first tag ribosome densities; the first translation efficiency tag vector consists of a plurality of first tag translation efficiencies; the first translation elongation rate tag vector consists of a plurality of first tag translation elongation rates; the first translation blocking point tag vector consists of a plurality of first tag translation point types; the first tag translation point types include blocking points and non-blocking points; the first tag footprint codes, the first tag ribosome density, the first tag translation efficiency, the first tag translation elongation rate and the first tag translation point types all correspond one-to-one to the first tag base codes;

[0107] Step 22, performing model training on the multi-task processing model based on the first data set;

[0108] Specifically, it includes: step 221, randomly dividing the first data set into two sub-data sets based on a preset first division ratio and recording them as a corresponding first training set and a first evaluation set;

[0109] The first segmentation ratio is a preset ratio value, such as 8:2; the first training set and the first evaluation set are both composed of a plurality of first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio;

[0110] Step 222, extracting the first first data record of the first training set as the corresponding current training record;

[0111] Step 223, vector conversion is performed on the first mRNA label sequence and the first Ribo-seq label sequence of the current training record to obtain the corresponding first training mRNA vector and the first training Ribo-seq vector; and the first ribosome density label vector, the first translation efficiency label vector, the first translation elongation rate label vector and the first translation barrier point label vector of the current training record are recorded as the corresponding first ribosome density label vector PD tag , the first translation efficiency label vector PE tag , the first translation elongation rate label vector PV tag and the first translation obstruction point label vector PS tag ;

[0112] Step 224, calculate the mean of all first label footprint codes of the first training Ribo-seq vector to obtain the corresponding current average code; and use the largest and smallest first label footprint codes in the first training Ribo-seq vector as the corresponding current maximum code and current minimum code; and record the coding range from the current minimum code to the current average code and the coding range from the current average code to the current maximum code as the corresponding low-order coding range and high-order coding range; and randomly select an integer between the preset positive and negative integer ranges [-R1, R1] as the corresponding current random number R2; and when the current random number R2 is greater than 0, randomly select R2 first label footprint codes from the first training Ribo-seq vector as the corresponding random scrambling code object; and compare each random scrambling code object with the current average code, if the current random scrambling code object is less than or equal to the current average code, randomly select a code from the high-order coding range to modify the current random scrambling code object, if the current random scrambling code object is greater than the current average code, randomly select a code from the low-order coding range to modify the current random scrambling code object;

[0113] Here, the positive and negative integer range [-R1, R1] is a preset value range, where R1 is a preset positive integer;

[0114] In addition, the embodiment of the present invention provides a processing method for randomly scrambling the first training Ribo-seq vector in the current step; the model's tolerance to erroneous data can be improved through this random scrambling operation; the multi-task processing model trained based on this scrambling method can output a high-quality prediction result even if the Ribo-seq data input to the model has more technical interference;

[0115] Step 225, input the first training mRNA vector and the first training Ribo-seq vector into the multi-task processing model to perform ribosome density, translation efficiency, translation elongation rate and translation blocking point prediction processing to obtain the corresponding first ribosome density prediction vector PDpre , the first translation efficiency prediction vector PE pre , the first translation elongation rate prediction vector PV pre and the first translation obstacle prediction vector PS pre ;

[0116] Step 226, the first ribosome density prediction vector PD pre and the first ribosome density label vector PD tag Substitute the preset first loss function L D , and the first translation efficiency prediction vector PE pre and the first translation efficiency label vector PE tag Substitute the preset second loss function L E , and the first translation elongation rate prediction vector PV pre and the first translation elongation rate tag vector PV tag Bring in the preset third loss function L V , and the first translation barrier point prediction vector PS pre and the first translation obstruction point label vector PS tag Substitute the preset fourth loss function L S , and by the first loss function L D , the second loss function L E , the third loss function L V And the fourth loss function L S Add together to form the corresponding first model loss function L M ;

[0117] Among them, the first model loss function L M for:

[0118] L M =L D (PD pre ,PD tag )+L E (PE pre ,PE tag )+L V (PV pre ,PV tag )+L S (PS pre ,PS tag );

[0119] The first loss function L D , the second loss function L E And the third loss function L V At least including MSE loss function, MAE loss function, Huber loss function, quantile loss function and SmoothL1 loss function; the fourth loss function LS At least including cross entropy loss function, logarithmic loss function and hinge loss function;

[0120] Step 227, based on a preset first model optimizer and a preset first iteration threshold, moves toward making the first loss function L D , the second loss function L E , the third loss function L V , the fourth loss function L S And the first model loss function L M A round of iterative modulation is performed on the model parameters of the multi-task processing model in the direction in which the minimum value can be achieved;

[0121] Wherein, the first model optimizer includes at least an SGD optimizer and an ADAM optimizer; the first iteration threshold is a preset integer value;

[0122] Step 228, identifying whether the current training record is the last first data record of the first training set; if so, go to step 229; if not, extract the next first data record of the first training set as the new current training record and return to step 223;

[0123] Step 229, perform a round of traversal on all the first data records of the first evaluation set; and in this round of traversal, use the currently traversed first data record as the corresponding current evaluation record; and perform vector conversion on the first mRNA tag sequence and the first Ribo-seq tag sequence of the current evaluation record to obtain the corresponding second training mRNA vector and the second training Ribo-seq vector; and input the second training mRNA vector and the second training Ribo-seq vector into the multi-task processing model for ribosome density, translation efficiency, translation elongation rate and translation blocking point prediction processing to obtain the corresponding second ribosome density prediction vector, second translation efficiency prediction vector, second translation elongation rate prediction vector and second translation blocking point prediction vector; and form the second ribosome density prediction vector, the second translation efficiency prediction vector and the second translation elongation rate prediction vector The first prediction tensor is formed by the first ribosome density label vector, the first translation efficiency label vector and the first translation elongation rate label vector of the current training record to form a corresponding first label tensor, and the first translation barrier label vector of the current training record is used as the corresponding second label tensor; the first prediction tensor and the first label tensor form a corresponding first prediction-label pair, and the second prediction tensor and the second label tensor form a corresponding second prediction-label pair; at the end of this round of traversal, all the first prediction-label pairs obtained are brought into the preset first evaluation function to calculate the corresponding first evaluation value; and the precision, recall and F1 score evaluations are performed on all the second prediction-label pairs obtained to obtain the corresponding first precision, first recall and first F1 score;

[0124] Wherein, the first evaluation function includes at least an MSE function and an RMSE function;

[0125] Step 230, identify the first evaluation value, the first precision, the first recall rate and the first F1 score; if the first evaluation value, the first precision, the first recall rate or the first F1 score does not satisfy the corresponding evaluation value range, precision range, recall rate range or F1 score range, return to step 221 to continue training; if the first evaluation value, the first precision, the first recall rate and the first F1 score all satisfy their respective corresponding evaluation value range, precision range, recall rate range and F1 score range, stop training and confirm that the model training of the multi-task processing model is completed.

[0126] The evaluation value range, precision range, recall range or F1 score range here are four preset value ranges.

[0127] Step 3: After the model training is completed, the first mRNA sequence and the first Ribo-seq sequence input by the user are received.

[0128] Step 4, performing vector conversion on the first mRNA sequence and the first Ribo-seq sequence to obtain the corresponding first mRNA vector and the first Ribo-seq vector; and using the multi-task processing model to perform ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing based on the first mRNA vector and the first Ribo-seq vector to obtain the corresponding first ribosome density vector, first translation efficiency vector, first translation elongation rate vector and first translation barrier point vector.

[0129] Step 5, correcting the first Ribo-seq sequence according to the first ribosome density vector; and marking the ribosome high density region, efficient translation region, extension rate change region and potential blocking point of the first mRNA sequence according to the first ribosome density vector, the first translation efficiency vector, the first translation elongation rate vector and the first translation blocking point vector; and feeding back the corrected first Ribo-seq sequence and the marked first mRNA sequence to the user;

[0130] Specifically, it includes: step 51, correcting the first Ribo-seq sequence according to the first ribosome density vector;

[0131] Specifically, the method comprises: step 511, performing normalization calculation on all ribosome densities of the first ribosome density vector to obtain a corresponding first normalized sequence; and performing normalization calculation on all ribosome footprint count values ​​of the first Ribo-seq sequence to obtain a corresponding second normalized sequence;

[0132] Wherein, the first and second normalized sequences are each composed of a plurality of corresponding first and second normalized values; the first and second normalized values ​​correspond to each other one by one;

[0133] It should be noted here that the normalization algorithm and the corresponding denormalization algorithm used in the embodiment of the present invention when normalizing the first ribosome density vector and the first Ribo-seq sequence are the same, and the normalization algorithm / denormalization algorithm used in the embodiment of the present invention is a set of conventional normalization / denormalization algorithms, such as maximum-minimum normalization / denormalization algorithm, Z-score normalization / denormalization algorithm, decimal calibration normalization / denormalization algorithm, etc., and can also be customized according to actual application requirements;

[0134] Step 512, performing absolute difference calculation on the first and second normalized values ​​corresponding to each group to obtain a corresponding first absolute difference; and comparing each first absolute difference with a preset absolute difference threshold, if the currently compared first absolute difference exceeds the absolute difference threshold, recording the ribosome footprint count value corresponding to the current first absolute difference in the first Ribo-seq sequence as a corresponding count value to be corrected;

[0135] Here, the absolute difference threshold is a preset threshold parameter;

[0136] Step 513, identifying whether the total number of count values ​​to be corrected is 0;

[0137] Step 514: if the total number of count values ​​to be corrected is 0, the first Ribo-seq sequence without modification is directly output as the corrected first Ribo-seq sequence;

[0138] Step 515, if the total number of count values ​​to be corrected is greater than 0, then a denormalization calculation is performed according to the first normalized value corresponding to each count value to be corrected to obtain a corresponding corrected count value; and based on each corrected count value, the corresponding count value to be corrected in the first Ribo-seq sequence is reset; and a correction label is marked on each reset ribosome footprint count value on the first Ribo-seq sequence; and the modified and marked first Ribo-seq sequence is output as a corrected first Ribo-seq sequence;

[0139] Step 52, marking the ribosome high density region, the efficient translation region, the elongation rate change region and the potential blocking point of the first mRNA sequence according to the first ribosome density vector, the first translation efficiency vector, the first translation elongation rate vector and the first translation blocking point vector;

[0140] Specifically, it includes: step 521, recording all ribosome densities exceeding a preset first density threshold in the first ribosome density vector as corresponding first high densities; and marking all base unit fragments corresponding to two or more consecutive first high densities in the first mRNA sequence as corresponding ribosome high density regions;

[0141] Here, the first density threshold is a preset ribosome density reference value;

[0142] Step 522, all translation efficiencies exceeding a preset first efficiency threshold in the first translation efficiency vector are recorded as corresponding first high efficiencies; and all base unit fragments corresponding to two or more consecutive first high efficiencies in the first mRNA sequence are marked as corresponding high-efficiency translation regions;

[0143] Here, the first efficiency threshold is a preset translation efficiency reference value;

[0144] Step 523, performing differential calculation on two adjacent translation elongation rates in the first translation elongation rate vector to obtain corresponding first differential rates to form a corresponding first differential sequence; and recording the first differential rates in the first differential sequence that exceed a preset reasonable differential rate range as corresponding abnormal differential rates; and marking the base unit fragments consisting of two base units corresponding to each abnormal differential rate in the first mRNA sequence as a corresponding elongation rate change region;

[0145] Here, the reasonable differential rate range is a preset differential rate range;

[0146] It should be noted here that, in the embodiment of the present invention, the ratio of each first differential rate to the previous translation extension rate of the two corresponding translation extension rates can be used as the corresponding extension rate change rate, and based on another preset change rate threshold, the extension rate change rate exceeding the threshold is screened out from the obtained multiple extension rate change rates as the corresponding abnormal change rate, and the base unit fragments consisting of two base units corresponding to each abnormal change rate in the first mRNA sequence are all marked as a corresponding extension rate change region;

[0147] Step 524, all translation point types that are blocking points in the first translation blocking point vector are recorded as corresponding blocking point types; and all base units corresponding to each blocking point type in the first mRNA sequence are marked as corresponding potential blocking points;

[0148] Step 53, and feed back the corrected first Ribo-seq sequence and the labeled first mRNA sequence to the user.

[0149] Figure 3 The module structure diagram of a processing device for a multi-task processing model of mRNA provided in the second embodiment of the present invention, the device is a terminal device or server that implements the aforementioned method embodiment, and can also be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiment, for example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a model building module 201, a model training module 202, a data receiving module 203, a model prediction module 204 and an analysis output module 205.

[0150] The model building module 201 is used to build a multi-task processing model for predicting the ribosome density, translation efficiency, translation elongation rate and translation blocking points of mRNA sequences.

[0151] The model training module 202 is used to collect data on mRNA sequences of multiple species and corresponding Ribo-seq sequences and construct a first data set based on the collected data; and perform model training on the multi-task processing model based on the first data set; the mRNA sequence is composed of a plurality of base units arranged in sequence; the sequence length of the Ribo-seq sequence is consistent with the sequence length of the corresponding mRNA sequence; the Ribo-seq sequence is composed of a plurality of ribosome footprint count values ​​arranged in sequence, and the ribosome footprint count values ​​correspond one to one to the base units.

[0152] The data receiving module 203 is used to receive the first mRNA sequence and the first Ribo-seq sequence input by the user after the model training is completed.

[0153] The model prediction module 204 is used to perform vector conversion on the first mRNA sequence and the first Ribo-seq sequence to obtain the corresponding first mRNA vector and the first Ribo-seq vector; and the multi-task processing model performs ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing based on the first mRNA vector and the first Ribo-seq vector to obtain the corresponding first ribosome density vector, first translation efficiency vector, first translation elongation rate vector and first translation barrier point vector.

[0154] The analysis output module 205 is used to correct the first Ribo-seq sequence according to the first ribosome density vector; and mark the ribosome high density area, efficient translation area, extension rate change area and potential obstacle point of the first mRNA sequence according to the first ribosome density vector, the first translation efficiency vector, the first translation extension rate vector and the first translation obstacle point vector; and feed back the corrected first Ribo-seq sequence and the marked first mRNA sequence to the user.

[0155] An embodiment of the present invention provides a processing device for a multi-task processing model for mRNA, which can execute the method steps in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here.

[0156] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the model building module can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above-mentioned module is determined. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0157] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital signal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0158] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the above method embodiments are generated. The above computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) methods. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0159] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device may be a terminal device or a server for implementing the method of the aforementioned embodiment, or may be a terminal device or a server for implementing the method of the aforementioned embodiment connected to the aforementioned terminal device or server. Figure 4 As shown, the electronic device may include: a processor 301 (such as a CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transceiver 303. Various instructions may be stored in the memory 302 to complete various processing functions and implement the processing steps described in the aforementioned embodiment method. Preferably, the electronic device involved in the embodiment of the present invention also includes: a power supply 304, a system bus 305 and a communication port 306. The system bus 305 is used to realize the communication connection between components. The above-mentioned communication port 306 is used for connecting and communicating between the electronic device and other peripherals.

[0160] exist Figure 4The system bus 305 mentioned in the figure can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used to realize the communication between the database access device and other devices (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM), and may also include non-volatile memory (Non-Volatile Memory), such as at least one disk storage.

[0161] The above-mentioned processor can be a general-purpose processor, including a central processing unit CPU, a network processor (Network Processor, NP), a graphics processing unit (Graphics Processing Unit, GPU), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0162] It should be noted that an embodiment of the present invention further provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the method and processing process provided in the above embodiments.

[0163] The embodiment of the present invention provides a processing method, device, electronic device and computer-readable storage medium for a multi-task processing model for mRNA. As can be seen from the above content, the embodiment of the present invention constructs a multi-task processing model for predicting the ribosome density, translation efficiency, translation extension rate and translation blocking point of an mRNA sequence; and performs big data collection on multiple types of mRNA sequences of multiple species and corresponding Ribo-seq sequences, and trains the multi-task processing model based on the collected data; and after the training, the mRNA sequence and Ribo-seq sequence input by the user are converted into corresponding vectors and input into the multi-task processing model for processing to obtain the corresponding ribosome density, translation efficiency, translation extension rate and translation blocking point prediction vectors, and use the ribosome density prediction vectors to correct the Ribo-seq sequence input by the user, and use the obtained four types of prediction vectors to mark the mRNA sequence input by the user for ribosome high density area, efficient translation area, extension rate change area and potential blocking point. Through the implementation of the present invention, on the one hand, an end-to-end multi-task model can be directly used to predict the ribosome density, translation elongation rate and translation blocking points of mRNA, thereby shortening the data acquisition cycle and improving the data acquisition efficiency; on the other hand, the model prediction results can be directly used to correct the Ribo-seq data and analyze and mark the mRNA sequence, which not only improves the quality of Ribo-seq data but also adds convenient data analysis methods; on the other hand, the species generalization of the model is also improved through multi-species data training.

[0164] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0165] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A processing method for a multi-task processing model of mRNA, characterized in that: The method comprises: Construct a multi-task processing model for predicting ribosome density, translation efficiency, translation elongation rate and translation block points of mRNA sequences; Data collection is performed on the mRNA sequences and corresponding Ribo-seq sequences of multiple species and a first data set is constructed based on the collected data; and model training is performed on the multi-task processing model based on the first data set; the mRNA sequence is formed by sequentially sorting a plurality of base units; the sequence length of the Ribo-seq sequence is consistent with the sequence length of the corresponding mRNA sequence; the Ribo-seq sequence is formed by sequentially sorting a plurality of ribosome footprint count values, and the ribosome footprint count values ​​correspond to the base units one by one; After the model training is completed, the first mRNA sequence and the first Ribo-seq sequence input by the user are received; Performing vector conversion on the first mRNA sequence and the first Ribo-seq sequence to obtain a corresponding first mRNA vector and a first Ribo-seq vector; and performing ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing on the first mRNA vector and the first Ribo-seq vector by the multi-task processing model to obtain a corresponding first ribosome density vector, a first translation efficiency vector, a first translation elongation rate vector and a first translation barrier point vector; Correcting the first Ribo-seq sequence according to the first ribosome density vector; marking the first mRNA sequence with high ribosome density regions, efficient translation regions, extension rate variation regions, and potential blocking points according to the first ribosome density vector, the first translation efficiency vector, the first translation extension rate vector, and the first translation blocking point vector; and feeding back the corrected first Ribo-seq sequence and the marked first mRNA sequence to the user; The multi-task processing model is used to perform ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing according to the input mRNA vector A and Ribo-seq vector B, and output the corresponding ribosome density prediction vector PD, translation efficiency prediction vector PE, translation elongation rate prediction vector PV and translation barrier point prediction vector PS; The mRNA vector A corresponds to an mRNA sequence, and is encoded by multiple bases. i Sort by order, 1≤i≤N A , N A is the sequence length of the current mRNA sequence; the base encoding a i One-to-one correspondence with the base units of the current mRNA sequence; The Ribo-seq vector B corresponds to the Ribo-seq sequence of the current mRNA sequence, and is represented by N A Footprint code b i The footprint code b i One-to-one correspondence with the ribosome footprint count value of the current Ribo-seq sequence and with the base encoding a i One to one correspondence; The ribosome density prediction vector PD is composed of N A The ribosome density d i The ribosome density d i With the base coding a i One to one correspondence; The translation efficiency prediction vector PE is composed of N A Translation efficiency e i The translation efficiency e i With the base coding a i One to one correspondence; The translation elongation rate prediction vector PV is composed of N A The translation elongation rate v i The translation elongation rate v i With the base coding a i One to one correspondence; The translation obstacle point prediction vector PS is composed of N A Translation point types i The translation point type s i With the base coding a i One-to-one correspondence; the translation point type s i There are two types: obstruction points and non-obstruction points; The first model input end and the second model input end of the multi-task processing model are respectively used to receive the corresponding mRNA vector A and the Ribo-seq vector B, and the first model output end, the second model output end, the third model output end and the fourth model output end are respectively used to output the corresponding ribosome density prediction vector PD, the translation efficiency prediction vector PE, the translation elongation rate prediction vector PV and the translation blocking point prediction vector PS; The multi-task processing model includes a feature fusion module, a Uni-RNA model, a ribosome density prediction head, a translation efficiency prediction head, a translation elongation rate prediction head and a translation blocking point prediction head; The first and second input ends of the feature fusion module are respectively connected to the corresponding first and second model input ends, and the output end is connected to the input end of the Uni-RNA model; the output end of the Uni-RNA model is respectively connected to the input ends of the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation barrier point prediction head; the output ends of the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation barrier point prediction head are respectively connected to the corresponding first, second, third and fourth model output ends; The feature fusion module is used to encode the mRNA vector A and the Ribo-seq vector B respectively according to the encoding method of the input embedding layer corresponding to the Uni-RNA model to generate the corresponding first and second encoding vectors; and send the Hadamard product vector of the first and second encoding vectors to the Uni-RNA model as the corresponding fusion feature vector X; the fusion feature vector X is composed of N A Fusion features x i Arranged in order; The Uni-RNA model has completed pre-training; the Uni-RNA model is used to perform feature encoding processing on the fusion feature vector X to obtain a corresponding encoding feature vector Y and send it to the ribosome density prediction head, the translation efficiency prediction head, the translation elongation rate prediction head and the translation blocking point prediction head respectively; The ribosome density prediction head is implemented based on a type of nonlinear regression prediction model; the ribosome density prediction head is used to predict the ribosome density at each base unit position according to the encoding feature vector Y to obtain the corresponding ribosome density prediction vector PD; the nonlinear regression prediction model at least includes an MLP model, an NNR model, a GBDT model and an XGBoost model; The translation efficiency prediction head is implemented based on a type of nonlinear regression prediction model; the translation efficiency prediction head is used to predict the translation efficiency at each base unit position according to the encoding feature vector Y to obtain the corresponding translation efficiency prediction vector PE; The translation elongation rate prediction head is implemented based on a type of nonlinear regression prediction model; the translation elongation rate prediction head is used to predict the translation elongation rate at each base unit position according to the encoding feature vector Y to obtain the corresponding translation elongation rate prediction vector PV; The translation barrier point prediction head is implemented based on a binary classification model; the translation barrier point prediction head is used to perform binary classification identification on whether each base unit position point is a translation barrier point according to the encoding feature vector Y to obtain the corresponding translation barrier point prediction vector PS; the binary classification model at least includes a binary classifier implemented based on an MLP model, a binary classifier implemented based on a CNN model, a binary classifier implemented based on a decision tree model, and a binary classifier implemented based on a random forest model.

2. The processing method for the multi-task processing model of mRNA according to claim 1, characterized in that: The first data set covers multiple types of mRNA molecules of multiple species, and is composed of multiple first data records; each of the first data records corresponds to a type of mRNA molecules of a type of species; The first data record includes a first mRNA tag sequence, a first Ribo-seq tag sequence, a first ribosome density tag vector, a first translation efficiency tag vector, a first translation elongation rate tag vector, and a first translation blocking point tag vector; The first mRNA tag sequence is composed of multiple first tag base codes; the first Ribo-seq tag sequence is composed of multiple first tag footprint codes; the first ribosome density tag vector is composed of multiple first tag ribosome densities; the first translation efficiency tag vector is composed of multiple first tag translation efficiencies; the first translation elongation rate tag vector is composed of multiple first tag translation elongation rates; the first translation blocking point tag vector is composed of multiple first tag translation point types; the first tag translation point types include blocking points and non-blocking points; The first tag footprint code, the first tag ribosome density, the first tag translation efficiency, the first tag translation extension rate and the first tag translation point type all correspond one-to-one to the first tag base code.

3. The processing method for the multi-task processing model of mRNA according to claim 2, characterized in that: The performing model training on the multi-task processing model based on the first data set specifically includes: Step 41, based on a preset first segmentation ratio, randomly split the first data set into two sub-data sets recorded as a corresponding first training set and a first evaluation set; Wherein, both the first training set and the first evaluation set are composed of a plurality of the first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio; Step 42, extracting the first first data record of the first training set as the corresponding current training record; Step 43, performing vector conversion on the first mRNA label sequence and the first Ribo-seq label sequence of the current training record to obtain the corresponding first training mRNA vector and the first training Ribo-seq vector; and recording the first ribosome density label vector, the first translation efficiency label vector, the first translation elongation rate label vector and the first translation blocking point label vector of the current training record as the corresponding first ribosome density label vector PD tag , the first translation efficiency label vector PE tag , the first translation elongation rate label vector PV tag and the first translation obstruction point label vector PS tag ; Step 44, calculate the mean of all the first label footprint codes of the first training Ribo-seq vector to obtain the corresponding current mean code; and use the largest and smallest first label footprint codes in the first training Ribo-seq vector as the corresponding current maximum code and current minimum code; and record the coding range from the current minimum code to the current mean code, and the coding range from the current mean code to the current maximum code as the corresponding low-order coding range and high-order coding range; and randomly select an integer between the preset positive and negative integer range [-R1, R1] as the corresponding current A random number R2 is selected; and when the random number R2 is greater than 0, R2 first label footprint codes are randomly selected from the first training Ribo-seq vector as corresponding random scrambled code objects; and each random scrambled code object is compared with the current average code, if the current random scrambled code object is less than or equal to the current average code, a code is randomly selected from the high-order code range to modify the current random scrambled code object, and if the current random scrambled code object is greater than the current average code, a code is randomly selected from the low-order code range to modify the current random scrambled code object; Step 45, inputting the first training mRNA vector and the first training Ribo-seq vector into the multi-task processing model to perform ribosome density, translation efficiency, translation elongation rate and translation blocking point prediction processing to obtain the corresponding first ribosome density prediction vector PD pre , the first translation efficiency prediction vector PE pre , the first translation elongation rate prediction vector PV pre and the first translation obstacle prediction vector PS pre ; Step 46: transform the first ribosome density prediction vector PD pre and the first ribosome density label vector PD tag Substitute the preset first loss function L D , and the first translation efficiency prediction vector PE pre and the first translation efficiency label vector PE tag Substitute the preset second loss function L E , and the first translation elongation rate prediction vector PV pre and the first translation elongation rate tag vector PV tag Bring in the preset third loss function L V , and the first translation barrier point prediction vector PS pre and the first translation obstruction point label vector PS tag Substitute the preset fourth loss function L S , and by the first loss function L D 、The second loss function L E , the third loss function L V And the fourth loss function L S Add together to form the corresponding first model loss function L M ; Among them, the first model loss function L M for: L M =L D (PD pre ,PD tag )+L E (PE pre ,PE tag )+L V (PV pre ,PV tag )+L S (PS pre ,PS tag ); The first loss function L D 、The second loss function L E And the third loss function L V At least including MSE loss function, MAE loss function, Huber loss function, quantile loss function and Smooth L1 loss function; the fourth loss function L S At least including cross entropy loss function, logarithmic loss function and hinge loss function; Step 47, based on a preset first model optimizer and a preset first iteration threshold, moves toward making the first loss function L D 、The second loss function L E , the third loss function L V , the fourth loss function L S And the first model loss function L M Performing a round of iterative modulation on the model parameters of the multi-task processing model in a direction in which the minimum value can be achieved; Wherein, the first model optimizer includes at least an SGD optimizer and an ADAM optimizer; Step 48, identifying whether the current training record is the last first data record of the first training set; if so, going to step 49; if not, extracting the next first data record of the first training set as the new current training record and returning to step 43; Step 49, perform a round of traversal on all the first data records of the first evaluation set; and in this round of traversal, use the first data record currently traversed as the corresponding current evaluation record; and perform vector conversion on the first mRNA tag sequence and the first Ribo-seq tag sequence of the current evaluation record to obtain the corresponding second training mRNA vector and the second training Ribo-seq vector; and input the second training mRNA vector and the second training Ribo-seq vector into the multi-task processing model for ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing to obtain the corresponding second ribosome density prediction vector, second translation efficiency prediction vector, second translation elongation rate prediction vector and second translation barrier point prediction vector; and the second ribosome density prediction vector, the second translation efficiency prediction vector and the second translation elongation rate prediction vector are composed of The corresponding first prediction tensor is composed of the second translation barrier point prediction vector as the corresponding second prediction tensor; and the first ribosome density label vector, the first translation efficiency label vector and the first translation elongation rate label vector of the current training record form a corresponding first label tensor, and the first translation barrier point label vector of the current training record is used as the corresponding second label tensor; and the first prediction tensor and the first label tensor form a corresponding first prediction-label pair, and the second prediction tensor and the second label tensor form a corresponding second prediction-label pair; and at the end of this round of traversal, all the first prediction-label pairs obtained are brought into the preset first evaluation function for calculation to obtain the corresponding first evaluation value; and the precision, recall and F1 score evaluations are performed on all the second prediction-label pairs obtained to obtain the corresponding first precision, first recall and first F1 score; Wherein, the first evaluation function includes at least an MSE function and an RMSE function; Step 50, identifying the first evaluation value, the first precision, the first recall rate and the first F1 score; if the first evaluation value, the first precision, the first recall rate or the first F1 score does not satisfy the corresponding evaluation value range, precision range, recall rate range or F1 score range, returning to step 41 to continue training; if the first evaluation value, the first precision, the first recall rate and the first F1 score all satisfy their respective corresponding evaluation value range, precision range, recall rate range and F1 score range, stopping training and confirming that the model training of the multi-task processing model is completed.

4. The processing method for the multi-task processing model of mRNA according to claim 1, characterized in that: The correcting the first Ribo-seq sequence according to the first ribosome density vector specifically includes: Step 51, normalizing all ribosome densities of the first ribosome density vector to obtain a corresponding first normalized sequence; and normalizing all ribosome footprint count values ​​of the first Ribo-seq sequence to obtain a corresponding second normalized sequence; Wherein, the first and second normalized sequences are each composed of a plurality of corresponding first and second normalized values; the first and second normalized values ​​correspond to each other one by one; Step 52, performing absolute difference calculation on the first and second normalized values ​​corresponding to each group to obtain a corresponding first absolute difference; and comparing each of the first absolute differences with a preset absolute difference threshold, if the first absolute difference currently compared exceeds the absolute difference threshold, recording the ribosome footprint count value corresponding to the current first absolute difference in the first Ribo-seq sequence as a corresponding count value to be corrected; Step 53, identifying whether the total number of count values ​​to be corrected is 0; Step 54: if the total number of the count values ​​to be corrected is 0, the first Ribo-seq sequence without modification is directly output as the first Ribo-seq sequence after correction; Step 55, if the total number of the count values ​​to be corrected is greater than 0, then the corresponding corrected count values ​​are obtained by performing a denormalization calculation based on the first normalized values ​​corresponding to each of the count values ​​to be corrected; and the corresponding count values ​​to be corrected in the first Ribo-seq sequence are reset based on each of the corrected count values; and each reset ribosomal footprint count value is marked with a correction label on the first Ribo-seq sequence; and the modified and marked first Ribo-seq sequence is output as the corrected first Ribo-seq sequence.

5. The processing method for the multi-task processing model of mRNA according to claim 1, characterized in that: The marking of the first mRNA sequence with respect to the ribosome high density region, the efficient translation region, the elongation rate variation region and the potential blocking point according to the first ribosome density vector, the first translation efficiency vector, the first translation elongation rate vector and the first translation blocking point vector specifically includes: Step 61, recording all ribosome densities exceeding a preset first density threshold in the first ribosome density vector as corresponding first high densities; and marking all base unit fragments corresponding to two or more consecutive first high densities in the first mRNA sequence as corresponding ribosome high density regions; Step 62, recording the translation efficiencies exceeding the preset first efficiency threshold in the first translation efficiency vector as corresponding first high efficiencies; and marking the base unit fragments corresponding to two or more consecutive first high efficiencies in the first mRNA sequence as corresponding high-efficiency translation regions; Step 63, performing differential calculation on two adjacent translation elongation rates in the first translation elongation rate vector to obtain corresponding first differential rates to form a corresponding first differential sequence; and recording the first differential rates in the first differential sequence that exceed a preset reasonable differential rate range as corresponding abnormal differential rates; and marking the base unit fragments consisting of two base units corresponding to each of the abnormal differential rates in the first mRNA sequence as a corresponding extension rate change region; Step 64, record all translation point types that are blocking points in the first translation blocking point vector as corresponding blocking point types; and mark all base units corresponding to each blocking point type in the first mRNA sequence as a corresponding potential blocking point.

6. A device for executing the processing method of the multi-task processing model for mRNA according to any one of claims 1 to 5, characterized in that: The device comprises: a model building module, a model training module, a data receiving module, a model prediction module and an analysis output module; The model building module is used to build a multi-task processing model for predicting ribosome density, translation efficiency, translation elongation rate and translation blocking points of mRNA sequences; The model training module is used to collect data on the mRNA sequences and corresponding Ribo-seq sequences of multiple species and construct a first data set based on the collected data; and to perform model training on the multi-task processing model based on the first data set; the mRNA sequence is formed by sequentially sorting a plurality of base units; the sequence length of the Ribo-seq sequence is consistent with the sequence length of the corresponding mRNA sequence; the Ribo-seq sequence is formed by sequentially sorting a plurality of ribosome footprint count values, and the ribosome footprint count values ​​correspond to the base units one by one; The data receiving module is used to receive a first mRNA sequence and a first Ribo-seq sequence input by a user after the model training is completed; The model prediction module is used to perform vector conversion on the first mRNA sequence and the first Ribo-seq sequence to obtain the corresponding first mRNA vector and the first Ribo-seq vector; and the multi-task processing model performs ribosome density, translation efficiency, translation elongation rate and translation barrier point prediction processing on the first mRNA vector and the first Ribo-seq vector to obtain the corresponding first ribosome density vector, first translation efficiency vector, first translation elongation rate vector and first translation barrier point vector; The analysis output module is used to correct the first Ribo-seq sequence according to the first ribosome density vector; and mark the first mRNA sequence with high ribosome density areas, high-efficiency translation areas, extension rate variation areas and potential blocking points according to the first ribosome density vector, the first translation efficiency vector, the first translation extension rate vector and the first translation blocking point vector; and feed back the corrected first Ribo-seq sequence and the marked first mRNA sequence to the user.

7. An electronic device, characterized in that: include: memory, processors, and transceivers; The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method according to any one of claims 1 to 5; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a computer, the computer is enabled to execute the method according to any one of claims 1 to 5.

Images