[Paper Review] Deep Learning-Based Strategy for Macromolecules Classification with Imbalanced Data from Cellular Electron Cryotomography

Ziqian Luo, Xiangrui Zeng+2 more|arXiv (Cornell University)|Aug 27, 2019

Cell Image Analysis Techniques43 references4 citations

TL;DR

This paper proposes a multi-path convolutional neural network (CNN) with focal loss and mixup augmentation to address class imbalance in macromolecular complex classification from cellular cryo-electron tomography (Cryo-ET) data. By training specialized paths for each underrepresented class and combining them via ensemble learning, the method achieves superior performance on imbalanced datasets, marking the first deep learning application to this problem in computational biology.

ABSTRACT

Deep learning model trained by imbalanced data may not work satisfactorily since it could be determined by major classes and thus may ignore the classes with small amount of data. In this paper, we apply deep learning based imbalanced data classification for the first time to cellular macromolecular complexes captured by Cryo-electron tomography (Cryo-ET). We adopt a range of strategies to cope with imbalanced data, including data sampling, bagging, boosting, Genetic Programming based method and. Particularly, inspired from Inception 3D network, we propose a multi-path CNN model combining focal loss and mixup on the Cryo-ET dataset to expand the dataset, where each path had its best performance corresponding to each type of data and let the network learn the combinations of the paths to improve the classification performance. In addition, extensive experiments have been conducted to show our proposed method is flexible enough to cope with different number of classes by adjusting the number of paths in our multi-path model. To our knowledge, this work is the first application of deep learning methods of dealing with imbalanced data to the internal tissue classification of cell macromolecular complexes, which opened up a new path for cell classification in the field of computational biology.

Motivation & Objective

• To address the challenge of class imbalance in macromolecular complex classification from Cryo-ET datasets, where some classes have orders of magnitude fewer samples than others.
• To develop a flexible deep learning framework capable of handling varying numbers of classes in imbalanced biological imaging data.
• To apply and evaluate multiple imbalanced data strategies—such as data sampling, bagging, boosting, and genetic programming—within a novel CNN architecture.
• To demonstrate the effectiveness of combining focal loss and mixup augmentation in improving minority class recognition in 3D cellular tomograms.
• To establish a new benchmark for deep learning-based classification of macromolecular complexes in near-native cellular environments.

Proposed method

• Proposes a multi-path CNN architecture where each path is optimized for a specific macromolecular class, improving performance on underrepresented types.
• Integrates focal loss to reduce the influence of easy, dominant classes during training and focus learning on hard, rare examples.
• Applies mixup data augmentation to generate synthetic training samples by linear interpolation of input images and labels, enhancing generalization.
• Employs a binary classification strategy per path, where each path learns to distinguish one class from all others, then combines outputs via ensemble voting for multi-class prediction.
• Uses genetic programming to evolve classifiers that perform well on both minority and majority classes without prior data reweighting.
• Enables model adaptability by adjusting the number of paths to suit different numbers of target classes, increasing flexibility for diverse biological datasets.

Experimental results

Research questions

• RQ1 Can a multi-path CNN architecture with class-specific paths improve classification accuracy on imbalanced Cryo-ET datasets compared to standard deep learning models?
• RQ2 How effective are focal loss and mixup augmentation in mitigating the negative impact of class imbalance in 3D cellular tomography?
• RQ3 Can ensemble strategies like bagging, boosting, and genetic programming be effectively combined with a multi-path CNN to enhance minority class detection?
• RQ4 To what extent can the proposed model be generalized to different numbers of classes by adjusting the number of paths?
• RQ5 Is the proposed method the first successful application of deep learning with imbalanced data techniques to macromolecular complex classification in Cryo-ET?

Key findings

• The multi-path CNN with focal loss and mixup achieved state-of-the-art performance on imbalanced Cryo-ET data, significantly outperforming traditional classifiers like GSVM-RU.
• The model demonstrated high adaptability, maintaining strong performance across two-, three-, and four-class classification tasks by adjusting the number of paths.
• Experiments confirmed that combining focal loss with mixup augmentation led to the best overall classification accuracy, especially for rare macromolecular complexes.
• Genetic programming-based classifiers evolved to handle imbalanced data without data reweighting, showing robustness and generalization on both majority and minority classes.
• The use of path-specific binary classifiers allowed each network branch to specialize, leading to improved detection of individual macromolecular complexes.
• The study established the first deep learning framework for imbalanced classification in Cryo-ET, opening new avenues for computational biology and structural cell biology.