PhoTorch: a robust and generalized biochemical photosynthesis model fitting package based on PyTorch

Area of Science:

• Plant Physiology
• Computational Biology
• Artificial Intelligence

Background:

• Artificial intelligence (AI) has advanced plant phenotyping and predictive modeling.
• Opportunities remain in applying AI for parameter optimization in complex biophysical models.
• Accurate parameter fitting is crucial for understanding plant physiological processes.

Purpose of the Study:

• To develop novel software, PhoTorch, for fitting parameters of the Farquhar, von Caemmerer, and Berry (FvCB) biochemical photosynthesis model.
• To leverage AI advancements, specifically PyTorch, for efficient and robust parameter estimation.
• To provide a flexible and computationally efficient tool for analyzing photosynthetic data.

Main Methods:

• Developed PhoTorch software integrating PyTorch's AI framework for parameter optimization.
• Implemented algorithms for fitting both steady-state and non-steady-state gas exchange data.
• Enabled flexible fitting of temperature and light response parameters, including simultaneous fitting of light response and $$ curves.

Main Results:

• PhoTorch demonstrated high computational efficiency, being over four times faster than benchmark software.
• The software showed robustness in parameter estimation, accurately fitting $$ curves with variability and noise.
• PhoTorch offers flexibility in handling diverse response curves and sub-model functional forms.

Conclusions:

• PhoTorch provides a reliable and efficient tool for researchers analyzing photosynthetic data.
• The software's speed and accuracy are particularly beneficial for processing large datasets of non-steady-state $$ curves.
• PhoTorch enhances the application of AI in plant biophysical modeling and parameter optimization.