Prognosis prediction method and device for acute myeloid leukemia based on multi-omics fusion

Abstract

The application discloses a kind of based on multi-omics fusion's acute myeloid leukemia prognosis prediction method and device, the method includes: collection and pre-processing gene mutation data and gene expression data, to construct data set;Acute myeloid leukemia prognosis prediction model is constructed, and using data set is trained;The gene mutation data and gene expression data after pre-processing are respectively used as the data input of genomics and transcriptomics in trained prediction model, obtain the risk score of prognosis prediction;Wherein, two shared encoders in model share part parameter, two modal features extracted by two shared encoders are aligned based on CLIP feature, and the features extracted by private encoder and shared encoder of each modality are decoupled.The present application fully excavates the complementarity and synergy of genomics and transcriptomics data by hierarchical feature decoupling and dynamic fusion mechanism, realizes the improvement of acute myeloid leukemia patient prognosis prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of cancer prognosis prediction technology, and in particular to a method and device for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion. Background Technology

[0002] Acute myeloid leukemia (AML) is a highly heterogeneous hematopoietic malignancy with high morbidity and mortality. The prognosis of AML patients varies significantly; some patients can achieve long-term survival with treatment, while others cannot avoid relapse even with intensive therapy. Therefore, accurate prognostic prediction is crucial for optimizing individualized treatment plans and improving patient survival rates. First, AML prognostic prediction helps with risk stratification. By analyzing patients' molecular genetic characteristics, clinical indicators, and multimodal data, patients can be divided into different risk groups such as low-risk, intermediate-risk, and high-risk, thereby guiding treatment decisions. Second, accurate prognostic prediction can drive the discovery and validation of novel biomarkers. With the development of high-throughput sequencing technology and artificial intelligence methods, more and more gene mutations, epigenetic modifications, and multi-omics characteristics are being used in AML prognostic assessment. This not only improves the accuracy of prognostic models but also provides new targets for individualized treatment, promoting the development of precision medicine.

[0003] Traditional prognostic prediction relies primarily on single-omics data (such as gene mutations and chromosome karyotypes), but single-dimensional information often fails to fully reveal the complex mechanisms of AML. Multi-omics fusion strategies, by integrating data from multiple levels including genomics, transcriptomics, epigenetics, proteomics, and metabolomics, can provide a more comprehensive characterization of disease features, thereby improving the accuracy and reliability of AML prognostic prediction. First, multi-omics fusion can overcome the limitations of single-omics information. For example, while gene mutations play a crucial role in the occurrence and progression of AML, they cannot fully explain prognostic differences among patients. Combining transcriptomics data can further elucidate the regulatory mechanisms of gene expression, thus providing a more accurate risk assessment. Second, the fusion of multi-omics data helps discover new prognostic biomarkers. Traditional prognostic prediction models are mainly based on known genetic mutations and clinical features, while multi-omics analysis can uncover potential key pathways, molecular interaction networks, and disease subtypes. Furthermore, advanced machine learning and deep learning methods make the fusion of multi-omics data more efficient. By constructing multimodal learning models, feature interactions at different omics levels can be captured, thereby improving the robustness and generalization ability of prognostic prediction.

[0004] Key challenges in AML prognostic prediction based on multi-omics fusion are twofold: firstly, the significant differences between different omics data necessitate the design of omics-specific encoders for robust feature representation; secondly, considering potential feature conflicts or redundancy between different omics, addressing feature redundancy within omics is crucial while mapping different omics to the same feature semantic space. Furthermore, effectively fusing these multi-level and multi-scale feature information is also a factor to consider when designing fusion strategies. Therefore, obtaining robust representations of different omics data and performing feature alignment and fusion on heterogeneous features are essential for AML prognostic prediction research. Constructing a robust multi-omics fusion model can significantly improve the accuracy of AML prognostic prediction. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and device for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion.

[0006] The objective of this invention is achieved through the following technical solution: The first aspect of this invention provides a method for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion, specifically including:

[0007] Collect gene mutation data and gene expression data, and preprocess them to construct a dataset;

[0008] An acute myeloid leukemia prognostic prediction model was constructed to obtain a risk score for prognostic prediction. The prediction model includes a first private encoder, a first shared encoder, a second private encoder, a second shared encoder, a first intramodal fusion module, a second intramodal fusion module, an intermodal fusion module, a modal hybrid fusion module, and a risk prediction module.

[0009] The prediction model is trained using a dataset. During the training process, the parameters of the prediction model are adjusted with the goal of minimizing the total loss function of the prediction model in order to obtain a well-trained prediction model.

[0010] Preprocessed gene mutation data and gene expression data are used as genomics and transcriptomics data, respectively, and input into the trained prediction model to obtain risk scores for prognosis prediction.

[0011] Furthermore, the preprocessed gene mutation data is a sparse binary matrix, where each row represents a sample, each column represents a gene locus, and each value indicates whether the gene has mutated in the corresponding sample. If it has mutated, its value is 1; if it has not mutated, its value is 0.

[0012] The preprocessed gene expression data is a gene expression matrix, where each row represents a sample, each column represents a gene locus, and each value represents the expression level of the gene in the corresponding sample.

[0013] Furthermore, the process by which the prediction model obtains the risk score for prognostic prediction specifically includes:

[0014] Preprocessed gene mutation data and gene expression data are input into the prediction model as genomics and transcriptomics data, respectively. Genomics data are processed by a first private encoder and a first shared encoder to extract private and shared features. Transcriptomics data are processed by a second private encoder and a second shared encoder to extract private and shared features. The private and shared features of genomics are input into a first intra-modal fusion module to obtain a first fusion feature. The private and shared features of transcriptomics are input into a second intra-modal fusion module to obtain a second fusion feature. The shared features of genomics and transcriptomics are input into an inter-modal fusion module to obtain a third fusion feature. The first, second, and third fusion features are input into a modal hybrid fusion module to obtain a fourth fusion feature. The fourth fusion feature is input into a risk prediction module to obtain a prognostic risk score.

[0015] Furthermore, the first private encoder includes two one-dimensional convolutional layers, two max pooling layers, two fully connected layers, and an activation function. The one-dimensional convolutional layers are connected to the max pooling layers, the last max pooling layer is connected to the fully connected layer, the two fully connected layers are connected, and the activation function is connected after the last fully connected layer.

[0016] The second private encoder includes three fully connected layers and an activation function, wherein the three fully connected layers are connected in sequence, and the activation function is connected after the last fully connected layer;

[0017] Both the first shared encoder and the second shared encoder include a first fully connected layer, a first layer normalization, a second fully connected layer, a first activation function, a second layer normalization, a third fully connected layer, and a second activation function connected in sequence; wherein, the last two fully connected layers of the first shared encoder and the second shared encoder share parameters.

[0018] Furthermore, both the first intramodal fusion module and the second intramodal fusion module include two multi-head self-attention mechanism modules for private features and shared features, a fully connected layer, and a weighted fusion mechanism based on the fully connected layer;

[0019] The intermodal fusion module includes a feature splicing layer and a multilayer perceptron network connected thereto. The multilayer perceptron network includes two fully connected layers and an activation function. The two fully connected layers are connected together, and the activation function is connected after the last fully connected layer.

[0020] Furthermore, the modal hybrid fusion module includes three multi-head self-attention mechanism modules for the first fusion feature, the second fusion feature, and the third fusion feature, a first fully connected layer, a weighted fusion mechanism based on the first fully connected layer, a feature splicing layer, and a second fully connected layer.

[0021] Furthermore, the risk prediction module includes two fully connected layers and a ReLU activation function, a Sigmoid activation function, and a batch normalization function. The first fully connected layer is followed by the ReLU activation function, the ReLU activation function is followed by the batch normalization function, the batch normalization function is followed by the second fully connected layer, and the second fully connected layer is followed by the Sigmoid activation function.

[0022] Furthermore, training the prediction model using the dataset specifically includes:

[0023] Preprocessed gene mutation and gene expression data from the dataset are input into the prediction model as genomics and transcriptomics data, respectively. The orthogonality loss for genomics is calculated by decoupling private and shared features; the orthogonality loss for transcriptomics is calculated by decoupling private and shared features; the contrastive learning loss is calculated by aligning shared features from genomics and transcriptomics based on contrastive language-image pre-training; the likelihood loss function is calculated using the predicted risk score, corresponding survival status label, and survival time; the total loss function of the prediction model is calculated based on the orthogonality loss from genomics, the orthogonality loss from transcriptomics, the contrastive learning loss, and the likelihood loss function; the parameters of the prediction model are adjusted to minimize the total loss function until a preset number of training epochs are reached to obtain a well-trained prediction model.

[0024] A second aspect of the present invention provides a prognostic prediction device for acute myeloid leukemia based on multi-omics fusion, comprising one or more processors and a memory, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-described prognostic prediction method for acute myeloid leukemia based on multi-omics fusion.

[0025] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, is used to implement the above-described method for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion.

[0026] The beneficial effects of this invention are: In the task of predicting the prognosis of acute myeloid leukemia, this invention can make full use of multi-omics data such as genomics and transcriptomics, and adopt a feature decoupling and hierarchical fusion mechanism to deeply explore the complementarity and synergy of genomics and transcriptomics data at multiple levels of information. By effectively integrating the two omics data, the accuracy and stability of the prognosis prediction for patients with acute myeloid leukemia (AML) can be improved. Attached Figure Description

[0027] Figure 1 This is a flowchart of the prognostic prediction method for acute myeloid leukemia based on multi-omics fusion of the present invention;

[0028] Figure 2 This is a flowchart illustrating the architecture of the acute myeloid leukemia prognostic prediction model of the present invention.

[0029] Figure 3 This is a schematic diagram of the structure of the first proprietary encoder for gene mutation data according to the present invention;

[0030] Figure 4 This is a schematic diagram of the structure of the second proprietary encoder for gene expression data according to the present invention;

[0031] Figure 5 This is a schematic diagram of the structure of the first shared encoder and the second shared encoder of the present invention;

[0032] Figure 6 This is a schematic diagram of the structure of the first modal intra-modal fusion module and the second modal intra-modal fusion module of the present invention;

[0033] Figure 7 This is a schematic diagram of the intermodal fusion module of the present invention;

[0034] Figure 8 This is a schematic diagram of the modal hybridization module of the present invention;

[0035] Figure 9 This is a schematic diagram of the risk prediction module of the present invention;

[0036] Figure 10 This is a schematic diagram of a prognostic prediction device for acute myeloid leukemia based on multi-omics fusion according to the present invention. Detailed Implementation

[0037] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. It is obvious that the drawings used in the following description are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0038] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0039] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0040] The present invention will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0041] See Figure 1 The prognostic prediction method for acute myeloid leukemia based on multi-omics fusion of the present invention specifically includes the following steps:

[0042] (1) Collect gene mutation data and gene expression data, and preprocess them to construct a dataset.

[0043] Furthermore, the preprocessed gene mutation data It is a sparse binary matrix where each row represents a sample (individual), each column represents a gene locus, and each value indicates whether the gene is mutated in the corresponding sample; if mutated, the value is 1, and if not mutated, the value is 0. Preprocessed gene expression data. It is a gene expression matrix, where each row represents a sample (individual), each column represents a gene locus, and each value represents the expression level of that gene in the corresponding sample. Here, n represents the total number of samples. This represents the total number of gene loci in the gene mutation data. This represents the total number of gene loci in the gene expression data.

[0044] The preprocessing for gene mutation data specifically includes: first, excluding genes with a mutation frequency of less than 1% from the gene mutation data; then, labeling the remaining genes with mutations: if a gene site is mutated, it is labeled as 1; if a gene site is not mutated, it is labeled as 0. This is a very common preprocessing method, based on the data type required by the model input, and will not be elaborated further here.

[0045] Preprocessing of gene expression data includes: acquiring gene expression data via RNA-seq and standardizing it using high-throughput sequencing fragment count / million mapped reads per kilobase transcript (Htseq-FPKM); further processing involves converting FPKM values ​​to TPM (transcripts per million transcripts), adding 1 to the TPM values, and then applying a logarithmic transformation to stabilize variance and improve the reliability of statistical analysis; next, calculating the variance of gene expression, selecting the top 100 genes with the largest variance, and standardizing their expression levels. This is a common preprocessing method, based on the data type required by the model input, and will not be elaborated further here.

[0046] It should be understood that a dataset constructed based on preprocessed gene mutation data and gene expression data can be used for subsequent model training.

[0047] (2) Construct a prognostic prediction model for acute myeloid leukemia to obtain a risk score for prognostic prediction. This prediction model includes a first private encoder, a first shared encoder, a second private encoder, a second shared encoder, a first intramodal fusion module, a second intramodal fusion module, an intermodal fusion module, a modal hybrid fusion module, and a risk prediction module, such as... Figure 2 As shown.

[0048] Furthermore, the process by which the prediction model obtains the risk score for prognostic prediction specifically includes: inputting preprocessed gene mutation data and gene expression data into the prediction model as genomics and transcriptomics data, respectively; passing the genomics data through a first private encoder and a first shared encoder to extract private and shared features of the genomics; passing the transcriptomics data through a second private encoder and a second shared encoder to extract private and shared features of the transcriptomics; inputting the private and shared features of the genomics into a first intra-modal fusion module to obtain a first fusion feature; inputting the private and shared features of the transcriptomics into a second intra-modal fusion module to obtain a second fusion feature; inputting the shared features of the genomics and the shared features of the transcriptomics into an inter-modal fusion module to obtain a third fusion feature; inputting the first, second, and third fusion features into a modal hybrid fusion module to obtain a fourth fusion feature; and inputting the fourth fusion feature into a risk prediction module to obtain the risk score for prognostic prediction.

[0049] Furthermore, since gene mutation data is typically high-dimensional, sparse, and contains complex nonlinear relationships, it is necessary to design an encoder that can efficiently extract features while handling sparsity as the first private encoder. For example... Figure 3 As shown, the first private encoder mainly uses two one-dimensional convolutional layers and two max-pooling layers: a one-dimensional convolutional neural network (CNN) is used to capture useful patterns in gene mutation data through local feature extraction, i.e., the 1D convolutional layers are used to extract the local relationships between gene mutations at adjacent gene sites; the max-pooling layers are used to reduce dimensionality and computational cost. Finally, two fully connected layers and one activation function are used, and the ReLU activation function can be selected.

[0050] Furthermore, considering that gene expression data typically involves a large number of genes and a limited number of samples, it is necessary to design an encoder capable of extracting useful features from high-dimensional, sparse data as a second private encoder. For example... Figure 4 As shown, in the second private encoder, since gene expression data often exhibits high-dimensional and complex nonlinear relationships, a variational autoencoder (VAE) can be used to better model the latent space. It mainly consists of three fully connected layers and one activation function, which can be ReLU, Leaky ReLU, or other types of activation functions. The first fully connected layer maps the input data to a high-dimensional hidden space, and then the latter two fully connected layers output the mean and log-variance of the latent space. A reparameterization technique is used to transform the mean and log-variance into latent space representations of the samples.

[0051] Furthermore, both the first and second shared encoders include three fully connected layers, two layer normalization functions, and two activation functions, such as... Figure 5 As shown, the first fully connected layer, the first layer normalization, the second fully connected layer, the first activation function, the second layer normalization, the third fully connected layer, and the second activation function are sequentially connected. The last two fully connected layers of the first shared encoder and the second shared encoder share parameters.

[0052] Furthermore, such as Figure 6 As shown, both the first and second intramodal fusion modules include two multi-head self-attention mechanism modules targeting private and shared features, a fully connected layer, and a weighted fusion mechanism based on the fully connected layer. One multi-head self-attention mechanism module targets private features, and the other targets shared features. The fully connected layer is used to learn weighting coefficients. Through training, the model automatically learns how to allocate weights, and then uses the weighting coefficients learned by the fully connected layer to perform a weighted summation of private and shared features to obtain the corresponding fusion feature. The first intramodal fusion module targets the private and shared features of genomics and can obtain the corresponding first fusion feature; the second intramodal fusion module targets the private and shared features of transcriptomics and can obtain the corresponding second fusion feature.

[0053] Furthermore, the intermodal fusion module includes a feature stitching layer and a connected multilayer perceptron network, such as... Figure 7 As shown, in the intermodal fusion module, to preserve all information from each modality, two aligned features—shared features from genomics and shared features from transcriptomics—are first concatenated. Then, a multilayer perceptron network is used to nonlinearly combine the concatenated features, effectively learning the complex relationships between features and performing feature fusion to obtain a third fused feature. This multilayer perceptron network consists of two fully connected layers and an activation function. The two fully connected layers are connected together, and the activation function is connected after the last fully connected layer.

[0054] Furthermore, such as Figure 8 As shown, the modality hybrid fusion module includes three multi-head self-attention mechanism modules for the first fusion feature, the second fusion feature, and the third fusion feature. Then, a fully connected layer learns the weight coefficients corresponding to the first fusion feature, the second fusion feature, and the third fusion feature. The first fusion feature, the second fusion feature, and the third fusion feature are weighted and summed using the weight coefficients learned by the fully connected layer. The weighted summed feature is then concatenated with the first fusion feature, the second fusion feature, and the third fusion feature processed by the corresponding multi-head self-attention mechanism modules. After concatenation, these features are mapped to the final latent feature space through a fully connected layer to obtain the fourth fusion feature.

[0055] Furthermore, such as Figure 9As shown, the risk prediction module includes two fully connected layers, a ReLU activation function, a Sigmoid activation function, and a batch normalization function (BatchNorm). The first fully connected layer is followed by the ReLU activation function, which is then followed by the batch normalization function. The batch normalization function is followed by the second fully connected layer, and the second fully connected layer is followed by the Sigmoid activation function. Specifically, the fourth fusion feature output from the modality mixing and fusion module is fed into the risk prediction module, passing sequentially through the first fully connected layer, the ReLU activation function, the batch normalization function, the second fully connected layer, and the Sigmoid activation function to obtain the predicted risk score. Batch normalization helps accelerate training and improve the model's generalization ability; the Sigmoid activation function is used in the output layer to control the risk score within a specific range (between 0 and 1).

[0056] (3) Use the dataset to train the prediction model. During the training process, the parameters of the prediction model are adjusted with the goal of minimizing the total loss function of the prediction model in order to obtain a well-trained prediction model.

[0057] Specifically, the preprocessed gene mutation data and gene expression data from the dataset are input into the prediction model as genomics and transcriptomics data, respectively. The orthogonality loss for genomics is calculated by decoupling the private and shared features of genomics; the orthogonality loss for transcriptomics is calculated by decoupling the private and shared features of transcriptomics; the contrastive learning loss is calculated by aligning the shared features of genomics and transcriptomics based on contrastive language-image pre-training; the likelihood loss function is calculated using the predicted risk score, corresponding survival status label, and survival time; the total loss function of the prediction model is calculated based on the orthogonality loss of genomics, the orthogonality loss of transcriptomics, the contrastive learning loss, and the likelihood loss function; the parameters of the prediction model are adjusted with the goal of minimizing the total loss function of the prediction model until a preset number of training epochs are reached to obtain a well-trained prediction model.

[0058] Furthermore, feature decoupling is required for the private and shared features extracted by the private and shared encoders corresponding to each modality. Specifically, feature decoupling is performed on the private and shared features extracted by the first private and first shared encoders for genomics, and on the private and shared features extracted by the second private and second shared encoders for transcriptomics. Orthogonality loss is used to force the shared and private features to be orthogonal in the representation space, effectively avoiding information redundancy. Therefore, the calculation formulas for the orthogonality loss in genomics and transcriptomics are as follows:

[0059] ;

[0060] ;

[0061] In the formula, This represents the orthogonal loss in genomics. Indicating shared characteristics in genomics, This indicates the private nature of genomics; This represents the orthogonal loss in transcriptomics. This indicates shared characteristics in transcriptomics. This indicates private features in transcriptomics; the superscript T indicates the transpose of a matrix or vector. This represents the Frobenius norm.

[0062] Furthermore, feature alignment based on Contrastive Language-Image Pre-Training (CLIP) is required for the two modal features extracted by the two shared encoders. Specifically, CLIP-based feature alignment is performed on the shared genomic features extracted by the first shared encoder and the shared transcriptomic features extracted by the second shared encoder. This involves constructing positive and negative sample pairs. Positive sample pairs consist of gene mutation and gene expression data pairs from the same patient, while negative sample pairs consist of gene mutation and gene expression data pairs from different patients. By maximizing the similarity of matching positive gene mutation-gene expression pairs while minimizing the similarity of negative sample pairs, cross-modal semantic alignment is achieved. The loss function used is InfoNCE loss, which makes matched gene mutation-gene expression pairs closer in the feature space and unmatched pairs further apart. In other words, the contrastive learning loss only applies to shared features, forcing cross-modal shared feature alignment. The formula for calculating this contrastive learning loss is:

[0063] ;

[0064] In the formula, Indicates the contrast learning loss; Represents an exponential function; This indicates the similarity between shared features of two modalities, namely, the similarity between shared features of the genomics modality and shared features of the transcriptomics modality; Representing shared characteristics of genomic modalities; Indicates and Shared features of matched transcriptomics modalities, namely gene expression data in positive sample pairs, are shared features of the same sample under transcriptomics modalities. The shared features of the i-th transcriptomics modality (which may be negative samples or candidate samples including positive samples) are represented; k represents the number of candidate sample pairs (including positive and negative sample pairs); It represents the temperature coefficient. The smaller the value, the sharper the distribution and the more it emphasizes the distinguishability.

[0065] Furthermore, in model training, the loss function for the prognostic risk score, corresponding label survival status, and survival time of AML patients is the likelihood loss (CoxLoss) function: for each sample i, its survival time is... Deleted marker is (1 indicates death upon the occurrence of the event, 0 indicates survival upon censoring), the risk score for prognostic prediction is: , Indicates time All samples that remain in the risk set afterward Let be the risk score predicted for the prognosis of sample j, and N be the total number of samples. Then the likelihood loss function is... The calculation formula is:

[0066] ;

[0067] Furthermore, the formula for calculating the total loss function of the prediction model is as follows:

[0068] ;

[0069] In the formula, This represents the total loss function of the prediction model. , and are learnable parameters, representing the weight coefficients of contrastive learning loss, orthogonal loss in genomics, and orthogonal loss in transcriptomics, respectively.

[0070] (4) The preprocessed gene mutation data and gene expression data are used as genomics and transcriptomics data, respectively, and input into the trained prediction model to obtain the risk score for prognosis prediction.

[0071] Corresponding to the aforementioned embodiments of the prognostic prediction method for acute myeloid leukemia based on multi-omics fusion, the present invention also provides embodiments of the prognostic prediction device for acute myeloid leukemia based on multi-omics fusion.

[0072] See Figure 10 The present invention provides an acute myeloid leukemia prognosis prediction device based on multi-omics fusion, comprising one or more processors and a memory, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the acute myeloid leukemia prognosis prediction method based on multi-omics fusion in the above embodiments.

[0073] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0074] The embodiments of the multi-omics fusion-based acute myeloid leukemia prognostic prediction device of this invention can be applied to any device with data processing capabilities, such as a computer. The device embodiments can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of any data processing device loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 10 The diagram shown is a hardware structure diagram of any data processing-capable device, including the multi-omics fusion-based acute myeloid leukemia prognosis prediction device of this invention. (Except for...) Figure 10 In addition to the processor, memory, network interface, and non-volatile memory shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0075] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0076] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0077] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the multi-omics fusion-based prognostic prediction method for acute myeloid leukemia described in the above embodiments.

[0078] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0079] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A prognostic prediction method for acute myeloid leukemia based on multi-omics fusion, characterized in that, Specifically, it includes: Collect gene mutation data and gene expression data, and preprocess them to construct a dataset; An acute myeloid leukemia prognostic prediction model was constructed to obtain a risk score for prognostic prediction. The prediction model includes a first private encoder, a first shared encoder, a second private encoder, a second shared encoder, a first intramodal fusion module, a second intramodal fusion module, an intermodal fusion module, a modal hybrid fusion module, and a risk prediction module. The prediction model is trained using a dataset. During the training process, the parameters of the prediction model are adjusted with the goal of minimizing the total loss function of the prediction model in order to obtain a well-trained prediction model. The training of the prediction model using the dataset specifically includes: Preprocessed gene mutation and gene expression data from the dataset are input into the prediction model as genomics and transcriptomics data, respectively. The orthogonality loss for genomics is calculated by decoupling private and shared features; the orthogonality loss for transcriptomics is calculated by decoupling private and shared features; the contrastive learning loss is calculated by aligning shared features from genomics and transcriptomics based on contrastive language-image pre-training; the likelihood loss function is calculated using the predicted risk score, corresponding survival status label, and survival time; the total loss function of the prediction model is calculated based on the orthogonality loss from genomics, the orthogonality loss from transcriptomics, the contrastive learning loss, and the likelihood loss function; the parameters of the prediction model are adjusted to minimize the total loss function until a preset number of training epochs are reached to obtain a well-trained prediction model. Preprocessed gene mutation data and gene expression data are used as genomics and transcriptomics data, respectively, and input into the trained prediction model to obtain risk scores for prognosis prediction.

2. The method for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion according to claim 1, characterized in that, The preprocessed gene mutation data is a sparse binary matrix, where each row represents a sample, each column represents a gene locus, and each value indicates whether the gene has mutated in the corresponding sample. If it has mutated, its value is 1; if it has not mutated, its value is 0. The preprocessed gene expression data is a gene expression matrix, where each row represents a sample, each column represents a gene locus, and each value represents the expression level of the gene in the corresponding sample.

3. The method for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion according to claim 1, characterized in that, The process by which the prediction model obtains the risk score for prognostic prediction specifically includes: Preprocessed gene mutation data and gene expression data are input into the prediction model as genomics and transcriptomics data, respectively. Genomics data are processed by a first private encoder and a first shared encoder to extract private and shared features. Transcriptomics data are processed by a second private encoder and a second shared encoder to extract private and shared features. The private and shared features of genomics are input into a first intra-modal fusion module to obtain a first fusion feature. The private and shared features of transcriptomics are input into a second intra-modal fusion module to obtain a second fusion feature. The shared features of genomics and transcriptomics are input into an inter-modal fusion module to obtain a third fusion feature. The first, second, and third fusion features are input into a modal hybrid fusion module to obtain a fourth fusion feature. The fourth fusion feature is input into a risk prediction module to obtain a prognostic risk score.

4. The method for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion according to claim 3, characterized in that, The first private encoder includes two one-dimensional convolutional layers, two max pooling layers, two fully connected layers, and an activation function. The one-dimensional convolutional layers are connected to the max pooling layers, the last max pooling layer is connected to the fully connected layer, the two fully connected layers are connected, and the activation function is connected after the last fully connected layer. The second private encoder includes three fully connected layers and an activation function, wherein the three fully connected layers are connected in sequence, and the activation function is connected after the last fully connected layer; Both the first shared encoder and the second shared encoder include a first fully connected layer, a first layer normalization, a second fully connected layer, a first activation function, a second layer normalization, a third fully connected layer, and a second activation function, which are connected in sequence; wherein, the last two fully connected layers of the first shared encoder and the second shared encoder share parameters.

5. The method for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion according to claim 3, characterized in that, Both the first intramodal fusion module and the second intramodal fusion module include two multi-head self-attention mechanism modules for private features and shared features, a fully connected layer, and a weighted fusion mechanism based on the fully connected layer; The intermodal fusion module includes a feature splicing layer and a multilayer perceptron network connected thereto. The multilayer perceptron network includes two fully connected layers and an activation function. The two fully connected layers are connected together, and the activation function is connected after the last fully connected layer.

6. The method for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion according to claim 3, characterized in that, The modal hybrid fusion module includes three multi-head self-attention mechanism modules for the first fusion feature, the second fusion feature, and the third fusion feature, a first fully connected layer, a weighted fusion mechanism based on the first fully connected layer, a feature splicing layer, and a second fully connected layer.

7. The method for predicting the prognosis of acute myeloid leukemia based on multi-omics fusion according to claim 3, characterized in that, The risk prediction module includes two fully connected layers, a ReLU activation function, a Sigmoid activation function, and a batch normalization function. The first fully connected layer is followed by the ReLU activation function, the ReLU activation function is followed by the batch normalization function, the batch normalization function is followed by the second fully connected layer, and the second fully connected layer is followed by the Sigmoid activation function.

8. A prognostic prediction device for acute myeloid leukemia based on multi-omics fusion, comprising one or more processors and a memory, characterized in that, The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the prognostic prediction method for acute myeloid leukemia based on multi-omics fusion as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, is used to implement the prognostic prediction method for acute myeloid leukemia based on multi-omics fusion as described in any one of claims 1-7.

Images