This is an early access version, the complete PDF, HTML, and XML versions will be available soon.
A Machine Learning Framework Integrating DeepLabCut and SimBA for Quantifying Aggressive Behavior in Swimming Crab Portunus trituberculatus
by
Chuanlong Ding
1,
Yuanyuan Fu
1,2,
Zhiqiang Zhou
3,
Ting Chen
3,4,
Wuqiang Xia
3,4,
Weijiang Huang
3,
Bincai Zhou
3,
Jiameng Chen
3,
Cheng Zheng
3,4,
Chunlin Wang
1,4,
Changkao Mu
1,4,
Changjun Liu
3,4 and
Lei Liu
1,4,*
1
School of Marine Sciences, Ningbo University, Ningbo 315832, China
2
Ningbo Institute of Oceanography, Ningbo 315832, China
3
Aquatic Technology Promotion Station of Xiangshan County, Ningbo 315709, China
4
Experimental Base of Xiangshan (Ningbo University) Aquatic Seed Industry Innovation Research Institute, Ningbo 315709, China
*
Author to whom correspondence should be addressed.
Animals 2026, 16(10), 1555; https://doi.org/10.3390/ani16101555
Submission received: 17 April 2026
/
Revised: 15 May 2026
/
Accepted: 17 May 2026
/
Published: 20 May 2026
(This article belongs to the Special Issue Behavioral Mechanisms in Aquatic Animals: Advances in Aquaculture and Breeding)
Simple Summary
Swimming crabs are an important species for seafood farming, but they often fight with each other. These fights can cause injuries and reduce the number of healthy crabs, which is a problem for aquaculture producers. To better understand and manage crab fighting, scientists need a way to measure aggression that is fast, consistent, and not based on personal judgment. In this study, we created a computer program that watches videos of crabs and automatically detects when they are fighting. The program tracks the crabs’ movements and calculates a score. We call this score the Time-weighted Aggression Index. It tells us how aggressive a crab is. We tested this score and found that it matches well with traditional methods of measuring aggression. We also showed that crabs with higher scores actually fight more often when placed together. This automatic tool saves time and provides a reliable way to study crab aggression. It can help crab farmers reduce fighting in their tanks and could even be used in breeding programs to select less aggressive crabs. In the future, the same approach could be adapted to study fighting in other animals as well.
Abstract
The swimming crab Portunus trituberculatus is a commercially important species in aquaculture, and its aggressiveness strongly influences productivity, making quantitative assessment a priority. However, such an assessment is hindered by behavioral complexity and a lack of high-throughput approaches. Here, we present a machine learning framework for quantifying aggressive behavior using DeepLabCut-SIMBA (Simple Behavioral Analysis). By converting video data into quantifiable durations of aggression-related events, we established a Time-weighted Aggression Index (TAI) that enables standardized, high-throughput assessment. The TAI showed strong concordance with conventional methods, confirming its validity. To demonstrate its applicability, we examined the correlation between TAI and serotonin levels, revealing distinct physiology-behavior associations. We then stratified crabs into three aggression categories based on TAI and conducted pairwise fighting trials. The frequency of aggressive interactions differed significantly across group combinations, indicating that TAI-based grouping effectively predicts behavioral outcomes. This framework offers a reliable, scalable tool for behavioral phenotyping and practical value for selective breeding programs.
