Method for predicting activities of trna, trfs and trna halves in traditional chinese medicine based on artificial intelligence

Abstract

An artificial intelligence-based traditional Chinese medicine-derived tRNA, tRFs, and tRNA halves activity prediction method. The present prediction method can achieve accurate prediction of cellular activities of different traditional Chinese medicine-derived tRNAs, and has good robustness. A deep learning algorithm used in the present method possesses a strong automatic feature learning capability, for automatically learning important feature information from big data, thereby avoiding extensive manual feature selection. A traditional Chinese medicine-derived tRNA, tRFs, and tRNA halves activity prediction model constructed in the present method has excellent prediction performance and good applicability. Therefore, an innovative idea is provided for the research and development of new drugs of traditional Chinese medicine-derived tRNAs.

Description

TECHNICAL FIELD [ 0001] The present disclosure relates to the field of computer-aided drug screening, in particular to a method for predicting activities of tRNA, tRFs, tRNA halves in a traditional Chinese medicine based on artificial intelligence, which is suitable for predicting cell activity according to a tRNA sequence in the traditional Chinese medicine. BACKGROUND [ 0002] The traditional Chinese medicine is a human treasure, but the specific action mechanism and the effective components of the traditional Chinese medicine are still to be researched and elucidated. Recently, it has been reported that non-coding RNA, such as microRNAs, have different modulatory effects by targeting different aspects of RNA transcription or post-transcriptional processes in almost all eukaryotes. Lin Zhang et al (CELL RESEARCH 2012, 22, 107-126) suggested that exogenous plant microRNAs in food could be absorbed by the mammalian gut and delivered to various organs via the blood stream, thereby regulating expression of mammalian genes. The group Zhihong Jiang, MACAU UNIVERSITY OF SCIENCE AND TECHNOLOGYy, found that the Ginseng-derived tRNA has a protective effect on cardiomyocytes (CN 111419867B) and the Taxus chinensis-derived tRNA has an inhibitory effect on ovarian cancer (Mol. Ther. Nucleic. 2022, 27, 718-732). [ 0003] For the drug effect research of traditional Chinese medicine-derived RNA, firstly, the cell activity is verified. Cell activity refers to the proportion of healthy cells in a sample population. The detection of cell activity is necessary for observing the physiological state of cells during the experimental process, and is also an important index for evaluating the therapeutic effect of drugs. Common methods of cell activity detection include MTT and CCK8. The dehydrogenases in living cells reduce WST-8 in exogenous MTT and CCK-8 reagents to water-insoluble blue-violet crystalline formazan and high water-insoluble yellow formazan products (Formazan dye), respectively, whereas dead cells do not have such function. The characteristics of MTT testing are high sensitivity and cost-effectiveness. But formazan crystals are insoluble in water and need to be dissolved in dimethyl sulfoxide (DMSO) before detection. Therefore, the workload is large, the accuracy is poor, and the DMSO is harmful to cells and experimenters. In contrast, the CCK-8 method is convenient to use, has high detection sensitivity, and low toxicity. However, due to factors such as cell growth cycle, the testing of large numbers of samples often involves long experimental cycles and high manpower costs. Therefore, developing efficient methods for evaluating drug activity is of great significance. [ 0004] With the development of information technology, domestic and international research has begun to explore to build computational models for predicting RNA drug activity by utilizing machine learning and RNA characteristics, which faces the following problems: 1) the prediction on the activity of diverse RNA sequences currently results in a low accuracy; 2) the predicted results lack biological significance and cannot be systematically explained from the perspective of biological mechanisms of action; 3) traditional machine learning methods cannot automatically learn characteristic information from large datasets, requiring extensive manual characteristic selection. Therefore, in view of the limitations of these methods, the present disclosure provides a method for predicting a cellular activity of tRNA, tRFs, and tRNA halves in a traditional Chinese medicine based on a deep learning, which can predict the cellular activity of tRNA, tRFs, and tRNA halves quickly, efficiently, and systematically. SUMMARY [ 0005] The technical problems to be solved: In order to eliminate the defects existing in the prior art, the present disclosure discloses a method for predicting cell activities of tRNA, tRFs and tRNA halves in a traditional Chinese medicine based on a deep learning algorithm (DEEP LEARNING Algorithm, DL). In the prediction method, a traditional Chinese tRNA database is fully combined with an artificial intelligence deep learning algorithm, which establishes a method for predicting the cell activities of tRNA, tRFs and tRNA halves in the the traditional Chinese medicine efficiently and rapidly, and provides an innovative thought for the research and development on new medicines of tRNA in the traditional Chinese medicines. [ 0006] The technical solutions: Provided is a method for predicting cell activities of tRNA, tRFs and tRNA halves in a traditional Chinese medicine based on artificial intelligence. The method comprises following steps. In Step 1, tRNA of a target traditional Chinese medicine is extracted and sequenced; In Step 2,tRFs and tRNA halves sequences are designed according to a sequencing result of the tRNA, a modeling data samples based on the traditional Chinese medicine tRNA are constructed; In Step 3, the tRNA of the target traditional Chinese medicine is purified, modification types and sites on the tRNA are determined by an LC-MS / MS method, modified tRFs and tRNAhalves sequences are designed, the modification is designed based on modification types of natural tRNA in the traditional Chinese medicine, by taking a stability and a drug efficacy of the tRFs and tRNA halves as indexes to adjust and optimize the modification types and sites; In Step 4, cell activity experiments are performed on the modified tRFs and tRNA halves, a screening result is taken as a final model sample characteristic; In Step 5, a prediction model is constructed based on a deep learning algorithm, the screened and obtained tRFs and tRNA halves are taken as a characteristic sequence for a model training and learning based on a presence or absence of cell activity, and In Step 6, key parameters for the model are optimized through a grid searching algorithm and a cross verification to achieve an improvement of a model prediction performance. [ 0007] Step 1 includes following steps: the target Chinese medicinal materials are Taxus chinensis, Panax ginseng, Ganoderma lucidum, venenum bufonis, Panax notoginseng, Salvia miltiorrhiza, or Ginkgo biloba. Total RNA of the traditional Chinese medicines are extracted, small RNA with a length less than 200 mer are separated and enriched, the total tRNAs in a small RNA sample are separated by adopting a 6% polyacrylamide gel electrophoresis containing 8M urea, stained by a SYBR nucleic acid dye, and then gel separation strips are observed under blue light, gel regions containing the total tRNA are cut, the cut gels are placed into a 3000 MWCO dialysis bag, the gel is eluted by adopting a transverse electrophoresis, eluent is recycled, the total tRNA in the eluent is purified by adopting a TRIzol method, a total tRNA sample is taken, a cDNA library is established by adopting a TruSeq SMALL RNA SAMPLE Preparation Kit, sequencing is conducted by adopting an illuminea second generation sequencing platform, a quality control is performed on tRNA second generation sequencing data by adopting a bioinformatics method, each tRNA sequence and content are analyzed, and eventually sequence information of the tRNA contained in the target traditional Chinese medicinal materials is obtained. [ 0008] Step 2 includes following steps: the tRFs and tRNA halves sequences are designed according to sequence information of the tRNA contained in the target traditional Chinese medicine, which includes (1) 5’-t-half that refers to a sequence formed from a break point to a 5’ end of the tRNA after an arbitrary break in an anticodon loop of an arbitrary tRNA; (2) 3’-t-half that refers to a sequence formed from a break point to a 3’ end of the tRNA after an arbitrary break in the anticodon loop of an arbitrary tRNA; (3) 5’-tRF that refers to a sequence formed by a break from the 5’ end of an arbitrary tRNA to an arbitrary position before the anticodon loop; (4) 3’-tRF that refers to a sequence formed by a break from the 3’ end of an arbitrary tRNA to an arbitrary position before the anticodon loop; 20% of the modeling data samples are taken as a test set, and remaining data are taken as a training set for a model construction. [ 0009] Step 3 includes following steps: a biotinylated DNA probe is designed according to tRNA sequence information in the target traditional Chinese medicine, an affinity capture is performied on the tRNA of the target traditional Chinese medicine by utilizing a complementary pairing principle between a probe and the tRNA sequence of the target traditional Chinese medicine, finally a biotinylated DNA probe-tRNA complex is recycled and purifed by adopting streptavidin-modified micro-nanomagnetic beads; a restriction hydrolysis is performed on purified tRNA monomers by adopting RNA restriction endonucleases or RNA exonucleases, and tRNA fragments are formed; a qualitative analysis is performed on the obtained tRNA fragments by adopting an LC-MS / MS method; the tRNA modification types and sites of the target traditional Chinese medicinal materials are confirmed by analyzing mass spectra of MS1 and MS2, tRFs and tRNA halves sequences containing modified nucleotides are designed according to sequencing results of LC-MS / MS modified sequences, and the modeling data sample based on the tRNA in the traditional Chinese medicinal is constructed, the modification types include pseudouracil, dihydrouracil, a base single-methylation modification, a base double-methylation modification, a base-trimethyl modification and a nucleotide five-carbohydrate 2'-oxymethyl modification. [ 0010] Step 4 includes following steps: related cells are selected according to diseases treated by a selected single traditional Chinese medicine, a cell activity screening is performed by CCK8, the cell activity experiment is performed that: a band with the IC50<100 nM or less than 50 nM and an inhibition rate exceeding 50% is selected as a characteristic sequence for tumor cells ; a band with a cell activity of 100%±10% at 50 nM and significantly different from a model group after modeling with a corresponding stimulation below 50 nM are selected as characteristic sequences for normal cells, according to a correlation and difference analysis of the cell activities of a blank control group and an experiment group; the sequences screened and obtained from cell experiments are converted into characteristic vectors as an input for a deep learning. The sequences are divided into at least K long base strings, K>12, by a K-mers method, and a frequency of the base strings is calculated as a characteristic representation of the sequence while dividing classification information of the sequence into activity and inactivity according to the cell experiments, is included in a model input. The characteristic vectors and the corresponding classification information are learned through a deep learning neural network, and a model capable of predicting an activity of unknown sequences is obtained. [ 0011] Step 5 includes following steps: a Sequential model interface is selected, a model structure includes an input layer, a hidden layer, and an output layer, and the hidden layer includes a dense layer and a dropout layer; an input of the model is a characteristic vector for the sequence processed and obtained in Step 4 and sequence activity classification information; an output of the model is a binary classification result of 0 and 1, where 0 represents inactivity (IC50>100 nM) and 1 represents activity (IC50<100 nM); in the hidden layers of the model, values for the input layer are activated by a Rectified Linear Unit (ReLU) activation function and then passed to the dense layer. A formula of the activation function is: y=ReLU(Wx + b); where x denotes a value for input data, y denotes a value after a data activation, W denotes a weight matrix, and b denotes a bias. In the output layer of the model, the values of the hidden layer are activated by a Sigmoid activation function and then output as an eventual output result. A formula of the activation function is: z=sigmoid(W’y+b’), where y denotes the value after the activation from the hidden layer, z denotes the result output by the model, W’ denotes a transposed weight matrix, and b’ denotes a transposed bias. The output layer of the model is further interfaced with public biomedical databases, including Array Express, Gene Expression, and Omnibus (GEO). During a training process of the model, a learning process of the model is configured by adopting a compile module, parameters for the model are set that: an optimizer is set to a Root Mean Square prop (RMSprop), a metrics list is set to an accuracy, and a loss function is set to a binary_crossentropy. A formula for calculating the loss function is: Lh(x, z) = -^^[Xk log Zk + (1 - Xk) log(1 - Zk)] where Lh(x,z) denotes a difference between a , predicted value and a true value, that is, a loss, x denotes a true value corresponding to the sample, z denotes a predicted value corresponding to the sample, and d denotes the number of epochs. [ 0012] The step that key parameters for the model are optimized through a grid searching algorithm and a cross verification in Step 6 includes following steps: a parameter optimization range is set, the epoch number is [10, 50, 100, 200, 500], a batch size is [10, 32, 64, 128], a learning rate is [0.01, 0.001, 0.00001], a dropout rate is [0,0.2,0.5], the node number is [50, 100, 300, 500, 1000], the constructed 5*4*3*5 models are optimized through a grid searching algorithm; the prediction performance of the model is evaluated through a ten-fold cross validation mode and an evaluation index, a performance evaluation index includes a sensitivity SEN, a specificity SPE, an accuracy ACC, a Matthews correlation coefficient MCC and an area AUC under a Receiver Operating Characteristic (ROC) curve, the closer the sensitivity, the specificity and the accuracy are to 100%, the closer the Matthews correlation coefficient and the area under ROC curve are to 1, indicating a better prediction performance of the model; on the contrary, the closer the sensitivity, the specificity and the accuracy are to 0, the closer the Matthews correlation coefficient and the area under the ROC curve are to 0.5, indicating a poorer prediction performance of the models; [ 0013] SEN = TP TP + FN TN [ 0014] SPE =------- FP+TN [ 0015] ACC = TP + TN TP + FN + FP + TN [ 0016] MCC = __________TP x TN - FP x FN__________ 7 (TP + FN) x (TP + FP) x (TN + FN) x (TN + FN) [ 0017] Where TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative. [ 0018] The beneficial effects: 1. The prediction method is capable of accurately predicting the cellular activities of different traditional Chinese medicine tRNAs in the traditional Chinese medicine and has a good robustness. 2. The deep learning algorithm adopted in the method has a stronger automatic characteristic learning capability, is capable of automatically learning important characteristic information from huge data, avoiding a large number of extensive manual characteristic selections. 3. The model for predicting the activities of the tRNA, tRFs, and tRNA halves in the traditional Chinese medicine based on the artificial intelligence constructed by the method has a better predictive performance and a better applicability, thereby providing an innovative though for the the research and the development on the new medicines of tRNA in the traditional Chinese medicines. BRIEF DESCRIPTION OF THE DRAWINGS [ 0019] FIG. 1 illustrates a general flow chart of a method for predicting tRNA in a traditional Chinese medicine based on a deep learning of the present disclosure. [ 0020] FIG. 2 illustrates a structural schematic diagram of a model for predicting activities of tRNA, tRFs, tRNA halves in a traditional Chinese medicine in a method of the present disclosure. [ 0021] FIG. 3 illustrates a graph of a prediction performance of the method for predicting the tRNA in the traditional Chinese medicine based on the deep learning. DETAILED DESCRIPTION OF THE EMBODIMENTS [ 0022] In order to make the objectives, technical solutions and advantages of the present disclosure more apparent, the technical solutions of the present disclosure will be further described below with reference to the accompanying drawings. [ 0023] Example 1 [ 0024] Provided in a specific technical solution is a method for predicting activities of tRNA, tRFs and tRNA halves in a traditional Chinese medicine based on artificial intelligence. The method is in the following. [ 0025] 157 RNA sequences are obtained by separating Ganoderma lucidum, and the data with the number of 1058 are obtained by CCK8 in human ovarian cancer cells (A2780), airway epithelial cells (16 HBE) and Airway Smooth Muscle Cells (ASMC), respectively. All modeling sample data are randomly divided into a training set and a test set at a ratio of 80% to 20%. [ 0026] The tRNA data obtained from sequencing are subjected to a quality control comparison to obtain tRNA sequence information. Through cell activity experiments, statistical analysis is performed on cells in the experimental group and control group. For A2780, bands with IC50<100 nM are selected as characteristic sequences. For ASMC, bands with the cell activity of 100%±10% within 50 nM and the inhibition rate>30% within 50 nM after TGF-P1 modeling are selected as characteristic sequences. The sequences are further converted into characteristic vectors, and K-mers are used to divide the sequences into groups of three bases. The frequency of the base groups is calculated as the characteristic representation of the sequence. At the same time, the sequence activity classification information of the cell experiments is included into the sequence characteristics. [ 0027] In this solution, a virtual environment is created by Python 3.6 of Anaconda5.1, and a tRNA activity prediction model is constructed by means of Keras deep learning frames and machine learning sklearn packages. The model selects a Sequential model interface and constructs a two-class prediction model. The model structure includes an input layer, a hidden layer, and an output layer, the hidden layer includes a dense layer and a dropout layer (FIG. 2). In the hidden layer of the model, the value for the input layer is activated by Rectified Linear Unit(ReLU) activation function and then is transmitted into the Dense layer, and the formula of the activation function is as in the following. [ 0028] y=ReLU(Wx+b) [ 0029] Where x denotes the value for the input data, y denotes the value after the data activation, W denotes the weight matrix, and b denotes the bias [ 0030] In the output layer of the model, the value for the hidden layer is activated by a Sigmoid activation function and then is transmitted out as a final output result, and the formula of the activation function is in the following. [ 0031] z=sigmoid(W’y+b’) [ 0032] Where y denotes an activated value transmitted by the hidden layer, z denotes a result output from the model, W' denotes a transposed weight matrix, and b' denotes a transposed deviation. [ 0033] The output layer of the model is further interfaced with public biomedical databases including Array Express, Gene Expression, and Omnibus (GEO). [ 0034] During the training process of the model, the learning process of the model is configured by adopting the compile module, the parameters for the model are set that: the optimizer is set to Root Mean Square prop (RMSprop), the metrics list is set to accuracy, and the loss function is set to binary_crossentropy, a formula for calculating the loss function is in the following. [ 0035] Lh(x, z) = -^^=1 [Xk log zt + (1 - Xk) log(1 - Zk)] [ 0036] Where LH (x, z) denotes the difference between the predicted value and the real value (that is, loss), x denotes the real value corresponding to the sample, z denotes the predicted value corresponding to the sample, and d denotes the number of epochs. [ 0037] Then the characteristic vector data of the final processed and obtained sequence is taken as the input of the model, 80% is taken as the training set to train the model and 20% is taken as the test set to test the model performance. [ 0038] The method for predicting the activities of the tRNA, tRFs and tRNA halves in the traditional Chinese medicine based on the artificial intelligence according to claim 1, Step 6 comprises the following steps. [ 0039] A parameter optimization range is set, the epoch number is [10, 50, 100, 200, 500], the batch size is [10, 32, 64, 128], the learning rate is [0.01, 0.001, 0.00001], the dropout rate is [0, 0.2, 0.5], and the node number is [50, 100, 300, 500, 1000]. [ 0040] The prediction performance of the trained model is evaluated by using test set samples in modeling data, Sensitivity (SEN), Specificity (SPE), Accuracy (ACC), Matthews correlation coefficient ( MCC), the area under the Receiver Operating Characteristic (ROC) curve (AUC) and other relevant evaluation indexes are respectively adopted for performance evaluation. Further, the key parameters (epoch number, batch size, learning rate, dropout rate, and node number) of the constructed 900 (5^4^3^3^5) models are optimized by using a grid searching algorithm and a ten-fold cross validation, so that the models obtain the optimal prediction performance. [ 0041] SEN = TP TP + FN [ 0042] SPE = TN FP+TN [ 0043] ACC = TP+TN TP+FN+FP+TN [ 0044] MCC = TP x TN - FP x FN 7 (TP + FN) x (TP + FP) x (TN + FN) x (TN + FN) [ 0045] Where TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative. [ 0046] Eventually, through parameter optimization, the specific parameters for the optimal model are set to be 2 layers of hidden layers and 50 nodes of each layer, the dropout rate is set to be 0.5 to avoid over fitting of the model, the learning rate is 0.001, the batch size is 128, and the epoch number is 50. The performance of the optimal model is inspected by a test set, the prediction accuracy is 97.1%, the AUC is 0.989, the sensitivity is 97.4%, the specificity is 96.8%, the Matthews correlation coefficient is 0.942, and in comparison with most of RNA prediction models based on traditional machine learning in China and abroad, the model has a better prediction performance (FIG. 3 and Table 1). [ 0047] The above instance is merely one embodiment of the present disclosure, and the simple alternations, substitutions are also within the scope of the present disclosure. [ 0048] TABLE 1 prediction results of a method for predicting tRNA in the traditional Chinese medicine based on the deep learning Confusion matrix In practice, activity In practice, inactivity Prediction of activity 185 32 Prediction of inactivity 5 1058

Claims

1. A method for predicting activities of tRNA, tRFs and tRNA halves in a traditional Chinese medicine based on artificial intelligence, comprising following steps:Step 1, extracting tRNA of a target traditional Chinese medicine and sequencing;Step 2, designing tRFs and tRNA halves sequences according to a sequencing result of the tRNA, and constructing a modeling data sample based on the tRNA in the traditional Chinese medicine;Step 3, purifying the tRNA of the target traditional Chinese medicine, determining modification types and sites on the tRNA by an LC-MS / MS method, designing modified tRFs and tRNA halves sequences, wherein the modification is designed based on modification types of natural tRNA in the traditional Chinese medicine by taking a stability and a drug effect of the tRFs and tRNA halves as indexes to adjust and optimize the modification types and sites;Step 4, performing cell activity experiments on the modified tRFs and tRNA halves, and taking a screening result as a final model sample characteristic;Step 5, constructing a prediction model based on a deep learning algorithm, and taking the screened and obtained tRFs and tRNA halves as a characteristic sequence for a model training and learning based on a presence or absence of cell activity; andStep 6, optimizing key parameters for the model through a grid searching algorithm and a cross verification to achieve an improvement of a model prediction performance.

2. The method for predicting the activities of the tRNA, tRFs and tRNA halves in the traditional Chinese medicine based on the artificial intelligence according to claim 1, wherein Step 1 includes following steps: the target Chinese medicinal materials are Taxus chinensis, Panax ginseng, Ganoderma lucidum, venenum bufonis, Panax notoginseng, Salvia miltiorrhiza, orGinkgo biloba; extracting total RNA of the traditional Chinese medicine, separating and enriching small RNA with a length less than 200 mer, separating the total RNAs in a small RNA sample by adopting a 6% polyacrylamide gel electrophoresis containing 8 M urea, staining by a SYBR nucleic acid dye, and then observing gel separation strips under blue light, cutting gel regions containing the total tRNA; placing the cut gels into 3000 MWCO dialysis bags, eluting by adopting a transverse electrophoresis, and recycling eluent; purifying the total tRNA in the eluent by adopting a TRzol method, taking a total tRNA sample, establishing a cDNA library by adopting a TruSeq SMALL RNA SAMPLE prepration Kit, sequencing by adopting an illumine second-generation sequencing platform; performing a quality control on tRNA second-generation sequencing data by adopting a bioinformatics method, analyzing each tRNA sequence and content, and eventually obtaining sequence information of the tRNA contained in the target traditional Chinese medicine.

3. The method for predicting the activities of the tRNA, tRFs and tRNA halves in the traditional Chinese medicine based on the artificial intelligence according to claim 1, wherein Step 2 includes following steps: designing the tRFs and tRNA halves sequences according to sequence information of the tRNA contained in the target traditional Chinese medicine, including (1) 5’-t-half: referring to a sequence formed from a break point to a 5’ end of the tRNA after an arbitrary break in an anticodon loop of an arbitrary tRNA; (2) 3’-t-half: referring to a sequence formed from a break point to a 3’ end of the tRNA after an arbitrary break in the anticodon loop of an arbitrary tRNA; (3) 5’-tRF: referring to a sequence formed by a break from the 5’ end of an arbitrary tRNA to an arbitrary position before the anticodon loop; (4) 3’-tRF: referring to a sequence formed by a break from the 3’ end of an arbitrary tRNA to an arbitrary position before the anticodon loop; taking 20% of the modeling data samples as a test set, and taking remaining data as a training set for a model construction.

4. The method for predicting the activities of the tRNA, tRFs and tRNA halves in the traditional Chinese medicine based on the artificial intelligence according to claim 1, wherein Step 3 includes following steps: designing a biotinylated DNA probe according to tRNA sequence information in the target traditional Chinese medicine, performing an affinity capture on thetRNA of the target traditional Chinese medicine by utilizing a complementary pairing principle between a probe and the tRNA sequence of the target traditional Chinese medicine, finally recycling and purifying a biotinylated DNA probe-tRNA complex by adopting streptavidin-modified micro-nanomagnetic beads, performing a restriction hydrolysis on purified tRNA monomers by adopting RNA restriction endonucleases or RNA exonucleases, forming tRNA fragments; performing a qualitative analysis on the obtained tRNA fragments by adopting an LC-MS / MS method; confirming the tRNA modification types and sites of the target traditional Chinese medicine by analyzing mass spectra of MS1 and MS2, designing tRFs and tRNA halves sequences containing modified nucleotides according to sequencing results of LC-MS / MS modified sequences, and constructing the modeling data sample based on the tRNA in the traditional Chinese medicine, wherein the modification types include pseudouracil, dihydrouracil, a base single-methylation modification, a base double-methylation modification, a base-trimethyl modification and a nucleotide five-carbohydrate 2' -oxymethyl modification.

5. The method for predicting the activities of the tRNA, tRFs and tRNA halves in the traditional Chinese medicine based on the artificial intelligence according to claim 1, wherein Step 4 includes following steps: selecting related cells according to diseases treated by a selected single traditional Chinese medicine, performing a cell activity screening by CCK8, performing the cell activity experiment as: selecting a band with IC50<100 nM or less than 50 nM and an inhibition rate exceeding 50% as a characteristic sequence for tumor cells; selecting a band with a cell activity of 100%±10% at 50 nM and significantly different from a model group after modeling with a corresponding stimulation below 50 nM as characteristic sequences for normal cells, according to a correlation and difference analysis of the cell activities of a blank control group and an experiment group; converting the sequences screened and obtained from cell experiments into characteristic vectors as an input for a deep learning, dividing the sequences into at least K long base strings, K^ 12, by a K-mers method and calculating a frequency of the base strings as a characteristic representation of the sequence while dividing classification information of the sequence into activity and inactivity according to the cell experiments and including in a model input; learning the characteristic vectors and the corresponding classification information through a deep learning neural network, obtaining a model capable of predicting an activity of unknown sequences.

6. The method for predicting the activities of the tRNA, tRFs and tRNA halves in the traditional Chinese medicine based on the artificial intelligence according to claim 1, wherein Step 5 includes following steps: selecting a Sequential model interface, wherein a model structure includes an input layer, a hidden layer and an output layer, the hidden layer includes a dense layer and a dropout layer; an input of the model is a characteristic vector for the sequence processed and obtained in Step 4 and sequence activity classification information; an output of the model is a binary classification result of 0 and 1, where 0 represents inactivity (IC50>100 nM) and 1 represents activity (IC50<100 nM); in the hidden layers of the model, values for the input layer are activated by a Rectified Linear Unit (ReLU) activation function and then passed to the dense layer, a formula of the activation function is:y=ReLU(Wx+b),where x denotes a value for input data, y denotes a value after a data activation, W denotes a weight matrix, and b denotes a bias;in the output layer of the model, the values for the hidden layer are activated by a Sigmoid activation function and then output as an eventual output result, a formula of the activation function is:z=sigmoid(W’y+b’);where y denotes the value after the activation from the hidden layer, z denotes the result output by the model, W’ denotes a transposed weight matrix, and b’ denotes a transposed bias;the output layer of the model is further interfaced with public biomedical databases, including Array Express, Gene Expression, and Omnibus (GEO);during a training process of the model, a learning process of the model is configured by adopting a compile module, parameters for the model are set that: an optimizer is set to a Root Mean Square prop (RMSprop), a metrics list is set to an accuracy, and a loss function is set to a binary_crossentropy, a formula for calculating the loss function is:Lh(x, z) = -^k=1 [Xk log Zk + (1 - Xk) log(1 - Zk)],where LH(x,z) denotes a difference between a predicted value and a true value, that is, a loss, x denotes a true value corresponding to the sample, z denotes a predicted value corresponding to the sample, and d denotes a number of epochs.7.The method for predicting the activities of the tRNA, tRFs and tRNA halves in the traditional Chinese medicine based on the artificial intelligence according to claim 1, wherein the step of optimizing key parameters for the model through a grid searching algorithm and a cross verification in Step 6 includes following steps: setting a parameter optimization range, wherein a epoch number is [10, 50, 100, 200, 500], a batch size is [10, 32, 64, 128], a learning rate is [0.01, 0.001, 0.00001], a dropout rate is [0, 0.2, 0.5], a node number is [50, 100, 300, 500, 1000]; optimizing the constructed 5X4X3X3X5 models through a grid searching algorithm; evaluating the prediction performance of the model through a ten-fold cross validation mode and an evaluation index, wherein a performance evaluation index includes a sensitivity SEN, a specificity SPE, an accuracy ACC, a Matthews correlation coefficient MCC, an area AUC under a ROC curve, the closer the sensitivity, the specificity and the accuracy are to 100%, the closer the Matthews correlation coefficient and the area under the ROC curve are to 1, indicating a better prediction performance of the model; on the contrary, the closer the sensitivity, the specificity and the accuracy are to 0, the closer the Matthews correlation coefficient and the area under the ROC curve are to 0.5, indicating a poorer prediction performance of the model;SEN =TPTP + FNSPE=TNFP + TNACC =TP + TNTP + FN + FP + TNMCC =__________TP x TN - FP x FN__________ 7 (TP + FN) x (TP + FP) x (TN + FN) x (TN + FN)where TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative.

Images