Diagnosis of Breast Hyperplasia and Evaluation of RuXian-I Based on Metabolomics Deep Belief Networks

Mingyang Jiang, Yanchun Liang, Zhili Pei, Xiye Wang, Fengfeng Zhou, Cheng‐Xi Wei, Xiaoyue FengInternational Journal of Molecular Sciences 2019 · DOI

TL;DR

This paper introduces a novel deep learning framework, the Metabolomics Deep Belief Network (MDBN), to accurately diagnose breast hyperplasia and evaluate the efficacy of the traditional Mongolian medicine RuXian-I using metabolomic data. The MDBN demonstrates superior classification performance compared to traditional methods, offering a promising avenue for rapid and precise assessment of therapeutic interventions in this context.

What Problem Does This Paper Address?

Breast cancer remains a significant health concern, and early diagnosis and treatment of its precursor, breast hyperplasia (BH), are crucial for prevention. While traditional Mongolian medicine like RuXian-I shows promise with high efficacy and low side effects, a rapid and accurate method to evaluate its effectiveness using metabolomic data has been lacking. Metabolomic data, derived from analyzing small molecules in biological samples, offers a detailed snapshot of an organism's physiological state. However, this data is often high-dimensional, sparse, and complex, posing challenges for traditional machine learning approaches. The authors aim to bridge this gap by developing a robust deep learning model capable of extracting meaningful patterns from metabolomic data for both diagnosis and treatment evaluation.

Methods

The core of this research is the Metabolomics Deep Belief Network (MDBN), a hybrid model combining Deep Belief Networks (DBNs) with softmax regression.

  • Deep Belief Networks (DBNs): DBNs are a type of deep neural network that uses Restricted Boltzmann Machines (RBMs) for unsupervised pre-training. This process allows the network to learn hierarchical representations of the data. The paper utilizes a two-layer RBM structure for pre-training, where the output of one RBM serves as the input for the next.
  • Softmax Regression: Following the DBN pre-training, a softmax regression classifier is employed for the final classification. This approach is an extension of logistic regression, well-suited for multi-class classification problems.
  • Training Strategy: The MDBN employs a two-stage training process:
  • Unsupervised Pre-training: Unlabeled metabolomic data is used to train the DBN layers, learning initial weights.
  • Supervised Fine-tuning: Labeled data is then used to fine-tune the entire DBN-softmax system. The limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) algorithm is used for this fine-tuning, known for its efficiency in optimization.
  • Dropout: To combat overfitting, a dropout strategy is incorporated during both pre-training and fine-tuning. This involves randomly setting a percentage of neuron outputs to zero during training, forcing the network to learn more robust features.
  • Data Preprocessing: Metabolomic data undergoes preprocessing including peak detection, alignment, and normalization. Both positive and negative ion modes from mass spectrometry are utilized to capture a comprehensive set of metabolites.

Datasets

The study utilizes metabolomic data collected from an animal model experiment involving female Wistar rats.

  • Total Samples: 168 samples were collected.
  • Groups: The rats were divided into three groups:
  • Control Group (CG)
    • Model Group (MG) - induced with oestradiol benzoate and progesterone to mimic breast hyperplasia.
  • RuXian-I Treatment Group (RG) - treated with RuXian-I.
  • Data Type: Serum metabolomic data was analyzed using Ultra-High-Performance Liquid Chromatography coupled with a quadrupole time-of-flight mass spectrometer (UPLC-Q-TOF MS).
  • Data Split: For evaluation, a five-fold cross-validation method was employed. Additionally, experiments were conducted with varying training and test set sizes to assess model performance under different data availability scenarios.

Key Insights and Ideas

  • Metabolomics as a Diagnostic Tool: The study strongly supports the utility of metabolomic data as a sensitive indicator for diagnosing breast hyperplasia and evaluating therapeutic responses. The distinct metabolic profiles across the control, model, and treatment groups highlight this.
  • DBN's Power in High-Dimensional Data: The research underscores the effectiveness of Deep Belief Networks in handling the inherent complexity of metabolomic data, which is characterized by high dimensionality and sparsity. DBNs excel at extracting abstract, high-level features that are not readily apparent with traditional methods.
  • L-BFGS for Efficient Optimization: The choice of the L-BFGS algorithm for fine-tuning proved significantly more efficient than gradient descent (GD), achieving comparable or better results with substantially fewer iterations and reduced training time. This is a crucial practical consideration for complex deep learning models.
  • Dropout's Role in Robustness: The implementation of dropout effectively mitigated overfitting, leading to more stable and reliable classification results, especially given the relatively small sample size.

Key Results

  • Superior Classification Accuracy: The MDBN model consistently outperformed traditional methods like BPNN, KNN, and SVM in classifying the metabolomic data.
  • For positive spectrum data, the MDBN achieved an average classification accuracy of 90.59% (DBN+GD+Softmax) and 95.30% (DBN+L-BFGS+Softmax), significantly higher than SVM (68.23%).
  • For negative spectrum data, the MDBN achieved an average accuracy of 92.94% (DBN+GD+Softmax) and 89.41% (DBN+L-BFGS+Softmax), outperforming SVM (78.23%).
  • Model Stability: The MDBN demonstrated stable classification results across different groups and when varying training/test set sizes. The L-BFGS fine-tuning method showed particularly stable performance.
  • Efficient Training: The DBN+L-BFGS+Softmax model achieved optimal performance with significantly fewer training epochs (around 200-500) compared to DBN+GD+Softmax (up to 5000 epochs), leading to reduced training time.
  • Clear Group Separation: The high classification accuracy indicated clear metabolic differences and boundaries between the control, model, and treatment groups, validating the animal model and the drug's efficacy.

Limitations and Future Work

  • Sample Size: While the study achieved good results, the authors acknowledge that metabolomic data can be characterized by small sample sizes. The use of cross-validation and dropout helped mitigate this, but larger datasets would further enhance model robustness.
  • Generalizability: The study was conducted on an animal model. Future work should focus on validating the MDBN framework on human clinical data to assess its generalizability to real-world diagnostic scenarios.
  • Feature Interpretation: While the MDBN excels at classification, further research could explore methods to interpret the specific metabolites driving these classifications, providing deeper biological insights into breast hyperplasia and RuXian-I's mechanism of action.
  • Broader Applications: The authors suggest that the MDBN framework can be extended to classify other types of metabolomic data and for different disease diagnoses, highlighting its potential as a versatile tool in biomedical research.

Summary updated

Abstract

Breast cancer is estimated to be the leading cancer type among new cases in American women. Core biopsy data have shown a close association between breast hyperplasia and breast cancer. The early diagnosis and treatment of breast hyperplasia are extremely important to prevent breast cancer. The Mongolian medicine RuXian-I is a traditional drug that has achieved a high level of efficacy and a low incidence of side effects in its clinical use. However, for detecting the efficacy of RuXian-I, a rapid and accurate evaluation method based on metabolomic data is still lacking. Therefore, we proposed a framework, named the metabolomics deep belief network (MDBN), to analyze breast hyperplasia metabolomic data. We obtained 168 samples of metabolomic data from an animal model experiment of RuXian-I, which were averaged from control groups, treatment groups, and model groups. In the process of training, unlabelled data were used to pretrain the Deep Belief Networks models, and then labelled data were used to complete fine-tuning based on a limited-memory Broyden Fletcher Goldfarb Shanno (L-BFGS) algorithm. To prevent overfitting, a dropout method was added to the pretraining and fine-tuning procedures. The experimental results showed that the proposed model is superior to other classical classification methods that are based on positive and negative spectra data. Further, the proposed model can be used as an extension of the classification method for metabolomic data. For the high accuracy of classification of the three groups, the model indicates obvious differences and boundaries between the three groups. It can be inferred that the animal model of RuXian-I is well established, which can lay a foundation for subsequent related experiments. This also shows that metabolomic data can be used as a means to verify the effectiveness of RuXian-I in the treatment of breast hyperplasia.
BibTeX
@article{jiang2019diagnosis,
  author = {Mingyang Jiang and Yanchun Liang and Zhili Pei and Xiye Wang and Fengfeng Zhou and Cheng‐Xi Wei and Xiaoyue Feng},
  title = {{Diagnosis of Breast Hyperplasia and Evaluation of RuXian-I Based on Metabolomics Deep Belief Networks}},
  journal = {International Journal of Molecular Sciences},
  year = {2019},
  doi = {10.3390/ijms20112620},
}
Figure