Multi-objective Genetic Programming with Multi-view Multi-level Feature for Enhanced Protein Secondary Structure Prediction

Yining Qian; Lijie Su; Meiling Xu; Xianpeng Wang

cs.LGMar 11, 2026

Multi-objective Genetic Programming with Multi-view Multi-level Feature for Enhanced Protein Secondary Structure Prediction

Authors: Yining Qian, Lijie Su, Meiling Xu, Xianpeng Wang

Organizations: National Frontiers Science Center for Industrial Intelligence and Systems Optimization, Shenyang 110819, China · Key Laboratory of Data Analytics and Optimization for Smart Industry (Northeastern University), Ministry of Education, Shenyang 110819, China · Liaoning Engineering Laboratory of Data Analytics and Optimization for Smart Industry, Shenyang 110819, China · Liaoning Key Laboratory of Manufacturing System and Logistics Optimization, Shenyang 110819, China ·

Abstract

Predicting protein secondary structure is essential for understanding protein function and advancing drug discovery. However, the intricate sequence-structure relationship poses significant challenges for accurate modeling. To address these, we propose MOGP-MMF, a multi-objective genetic programming framework that reformulates PSSP as an automated optimization task focused on feature selection and fusion. Specifically, MOGP-MMF introduces a multi-view multi-level representation strategy that integrates evolutionary, semantic, and newly introduced structural views to capture the comprehensive protein folding logic. Leveraging an enriched operator set, the framework evolves both linear and nonlinear fusion functions, effectively capturing high-order feature interactions while reducing fusion complexity. To resolve the accuracy-complexity trade-off, an improved multi-objective GP algorithm is developed, incorporating a knowledge transfer mechanism that utilizes prior evolutionary experience to guide the population toward global optima. Extensive experiments across seven benchmark datasets demonstrate that MOGP-MMF surpasses state-of-the-art methods, particularly in Q8 accuracy and structural integrity. Furthermore, MOGP-MMF generates a diverse set of non-dominated solutions, offering flexible model selection schemes for various practical application scenarios. The source code is available on GitHub: https://github.com/qian-ann/MOGP-MMF/tree/main.

Figures & tables

Fig. 1 : Crystal structure of chimeric carbonic anhydrase VI bound to ethoxzolamide (PDB ID: 6QL2), illustrating the protein secondary structure prediction.

Fig. 2 : The overall workflow of MOGP-MMF. It involves extracting multi-view features (PSSM, HMM, T5, SaProt), performing evolutionary feature fusion via an enriched operator set and genetic strategies, and finally selecting the optimal program for performance evaluation.

Fig. 3 : Architecture of multi-level feature extractors.

Function	Type	Description
$W\_Add$	linear	Add two weighted features
$W\_Sub$	linear	Subtract two weighted features
Mul	non-linear	Multiply filter
GRT	non-linear	Select maximum value between two features
Sqrt	non-linear	Sqrt an feature
Log	non-linear	Logarithm function

TABLE I : Functions in EOS

Constant	Type	Description
$n$	Integer	The parameters for the $W\_Add$ and $W\_Sub$ functions. They are randomly generated in the range of [1,10).
$\sigma$	Integer	The convolution kernel size of $LoGF$ is with values ranging from 1 to 4.
$FFTmode$	Integer	The transform dimension for $FFT$ with possible values of 1.

TABLE II : Ephemeral random constants in EOS

Hyperparameter	Setting
Number of epochs	100
Dropout rate	0.2
Batch size	64
Maximum sequence length	700
Number of labels	8
Kernel size (CNNs)	9

TABLE III : Hyperparameters used in GP-MMF

Dataset	Method	Q8	Precision	Recall	F1-score	MCC	Sov
CB6133	NetSurfP-3.0	0.765	0.771	0.765	0.756	0.711	76.2
	AttSec	0.783	0.774	0.783	0.773	0.730	78.9
	Mufold-SS	0.711	0.696	0.711	0.685	0.627	71.7
	ShuffleNet_SS	0.735	0.723	0.735	0.725	0.667	75.1
	DeepACLSTM	0.733	0.720	0.733	0.716	0.662	73.8
	2C-BLSTM	0.735	0.720	0.735	0.721	0.665	73.8

TABLE IV : Performance comparison of MOGP-MMF with state-of-the-art methods on benchmark datasets. The best results are highlighted in bold.

Method	Representation	CB6133	CB6133	CB513	CASP10	CASP11	CASP12	CASP13	CASP14
		(Validation)	(Test)
Single-view Single-level feature	HMM-CNN1	69.86	70.35	66.91	69.61	68.14	66.47	63.46	61.50
	HMM-CNN2	72.11	72.64	69.33	71.39	70.50	46.69	45.70	42.87
	HMM-RNN1	72.58	72.99	69.56	70.95	70.59	69.19	66.80	65.33
	HMM-RNN2	72.46	72.91	69.37	71.84	70.54	69.68	66.29	65.81
	PSSM-CNN1	69.44	69.58	66.33	68.91	66.24	59.70	58.63	59.53

TABLE V : Performance Comparison of MMF and Alternative Feature Configurations (Accuracy in Percentage)

Panel of Figure 9 — Fig. 4 : Visualization of comparisons between models with and without nonlinear operators on CB6133 dataset: (a) Accuracy comparison under MOGP, (b) Complexity comparison under MOGP.

Method	CB6133	CB6133	CB513	CASP10	CASP11	CASP12	CASP13	CASP14
	(Validation)	(Test)
Add	78.00	78.47	74.89	78.55	75.85	73.60	71.39	68.35
Mul	54.61	55.75	50.70	52.11	54.50	50.51	47.49	47.41
Max	77.97	78.25	74.75	78.55	75.62	72.08	69.59	67.79
Min	74.00	74.17	71.09	74.30	71.98	70.25	68.46	66.32
Concatenation	74.88	75.38	71.97	75.62	73.34	70.74	67.44	66.59

TABLE VI : Performance Comparison of MOGP and Alternative Fusion Methods (Accuracy in Percentage)

Panel of Figure 11 — Fig. 5 : Performance comparisons on the CB6133 dataset. (a-b) Accuracy and complexity of SOGP vs. Naive MOGP. (c-d) Impact of Knowledge Transfer (KT) on MOGP accuracy and complexity.

Fig. 6 : Representative ”anchor” trees from the 200-solution Pareto front (CB6133): (a) Low-Complexity, (b) Balanced, and (c) High-Performance.

Method	Anchor	Complexity	CB6133 (Test)	CB513	CASP10	CASP11	CASP12	CASP13	CASP14
Tree 1	Low-Complexity	15	79.39	75.98	79.76	76.41	68.28	66.15	62.70
Tree 2	Balanced	28	79.91	76.89	80.32	77.26	73.16	71.19	67.50
Tree 3	High-Performance	42	80.05	76.93	80.47	77.59	74.63	72.09	69.07

TABLE VII : Representative ”anchor” solution trees selected from the 200 non-dominated individuals in the 100th generation (CB6133 dataset): (a) Tree 1 (Low-Complexity), (b) Tree 2 (Balanced), and (c) Tree 3 (High-Performance). (Accuracy in Percentage)

Panel of Figure 14 — Fig. 7 : Visualization of feature representations across seven datasets. Rows (top to bottom): CASP10, CASP11, CASP12, CASP13, and CASP14. Columns: (a) Single-feature baseline, (b) Standard Multi-View Multi-Level Fusion (MMF), and (c) The proposed MOGP-MMF framework.

Multi-View Feature Fusion

See all 25 papers

Enhanced Protein Intrinsic Disorder Prediction Through Dual-View Multiscale Features and Multi-objective Evolutionary Algorithm
Mar 6, 2026Shaokuan Wang, Pengshan Cui, Yining Qian +2Multi-Scale Feature Fusion Protein Structure Prediction
College of Information Science and Engineering, Northeastern University, Shenyang 110819, China · School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China · Key Laboratory of Data Analytics and Optimization for Smart Industry, Ministry of Education, Northeastern University, Shenyang 110819, China
Protein Structure Prediction: From Evolutionary Constraints to Generative Modeling
Aug 17, 2026Wengan He, Yongsheng Luo, Lihong Jiang +2Protein Structure Prediction Protein Representation Learning
Zhuhai College of Science and Technology, Zhuhai, China · Zhuhai UNO Technology Co., Ltd., Zhuhai, China
Structure-aware Reinforcement Learning for Protein Directed Evolution
Sep 30, 2026Zikun Nie, Suyuan Zhao, Yizhen Luo +2Protein Design
Institute of AI Industry Research (AIR), Tsinghua University · Department of Computer Science and Technology, Tsinghua University · Pharmolix Inc.

Mar 6, 2026J/K move · Enter open · S save

Enhanced Protein Intrinsic Disorder Prediction Through Dual-View Multiscale Features and Multi-objective Evolutionary Algorithm

Shaokuan Wang, Pengshan Cui, Yining Qian, An-Yang Lu+1

College of Information Science and Engineering, Northeastern University, Shenyang 110819, China · School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China · Key Laboratory of Data Analytics and Optimization for Smart Industry, Ministry of Education, Northeastern University, Shenyang 110819, China

Multi-Scale Feature Fusion Protein Structure Prediction

Open paper AskPDF arXiv

Intrinsically disordered regions of proteins play a crucial role in cell signaling and drug discovery. However, their high structural flexibility makes accurate residue-level prediction challenging. Existing methods often rely on single-view representations or rigid manual fusion strategies, which fail to effectively balance the complex interplay between local amino acid preferences and long-range sequence patterns. To address these limitations, we propose D2MOE, a Dual-View Multiscale Features and Multi-objective Evolutionary Algorithm, which consists of two stages. First, a dual-view multiscale feature extraction method is introduced. This method integrates evolutionary views with deep semantic views and employs multiscale extractors to capture structural information across diverse receptive fields. Second, a multi-objective evolutionary algorithm is designed to adaptively discover optimal fusion architectures. By co-evolving discrete feature selection and continuous fusion weights, the algorithm adaptively explores optimal cross-feature architectures to enhance predictive accuracy while maintaining model compactness. Experimental results across three benchmark datasets demonstrate that D2MOE consistently outperforms state-of-the-art methods. D2MOE combines the feature extraction capabilities of deep learning with the global search advantages of evolutionary algorithms, enabling efficient feature integration without manual design, and providing a robust computational tool for protein disorder prediction.

Abstract

Multi-View Feature Fusion

Explore similar work

Enhanced Protein Intrinsic Disorder Prediction Through Dual-View Multiscale Features and Multi-objective Evolutionary Algorithm

Protein Structure Prediction: From Evolutionary Constraints to Generative Modeling

Structure-aware Reinforcement Learning for Protein Directed Evolution