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

TL;DR

PhoTorch addresses the challenge of estimating FvCB photosynthesis parameters from complex leaf gas exchange data, including non-steady-state measurements, by implementing the FvCB equations in PyTorch and optimizing parameters with automatic differentiation and the Adam algorithm. It supports fitting CO2, light, and temperature response parameters using flexible light and temperature functions, pre-processing to handle noisy data, and a loss with physiologically informed penalties. The tool demonstrates robust, efficient parameter recovery across field and synthetic data, including simultaneous A/Ci and A/Q fits, and shows favorable speed relative to benchmark packages while maintaining high fit quality. The open-source package enables direct integration with AI-based phenotyping pipelines and advances the ability to explore photosynthetic trait variation under diverse environmental conditions.

Abstract

Advancements in artificial intelligence (AI) have greatly benefited plant phenotyping and predictive modeling. However, unrealized opportunities exist in leveraging AI advancements in model parameter optimization for parameter fitting in complex biophysical models. This work developed novel software, PhoTorch, for fitting parameters of the Farquhar, von Caemmerer, and Berry (FvCB) biochemical photosynthesis model based the parameter optimization components of the popular AI framework PyTorch. The primary novelty of the software lies in its computational efficiency, robustness of parameter estimation, and flexibility in handling different types of response curves and sub-model functional forms. PhoTorch can fit both steady-state and non-steady-state gas exchange data with high efficiency and accuracy. Its flexibility allows for optional fitting of temperature and light response parameters, and can simultaneously fit light response curves and standard A/Ci curves. These features are not available within presently available A/Ci curve fitting packages. Results illustrated the robustness and efficiency of PhoTorch in fitting A/Ci curves with high variability and some level of artifacts and noise. PhoTorch is more than four times faster than benchmark software, which may be relevant when processing many non-steady-state A/Ci curves with hundreds of data points per curve. PhoTorch provides researchers from various fields with a reliable and efficient tool for analyzing photosynthetic data. The Python package is openly accessible from the repository: https://github.com/GEMINI-Breeding/photorch.

Figures (9)

• Figure 1: Example illustration of optional automated pre-processing of measured $A/C_i$ curves to smooth and remove artifacts at the beginning and end of the curves. The black and red points represent measured raw and pre-processed sample points, respectively.
• Figure 2: Results of $A/C_i$ fitting based on bean and cowpea field gas exchange data, excluding the light response curve. The horizontal and vertical coordinate of each point respectively corresponds to the modelled and measured $A$ values at each $C_i$ value. The learning rate and maximum iterations were set to 0.08 and 20,000 for all fittings.
• Figure 3: Fitted $V_{cmax25}$ versus $J_{max25}$ for each $A/C_i$ curve in the cowpea and bean datasets. Dashed lines give the best linear fit for each species.
• Figure 4: Correlation between temperature response parameters for each genotype in the bean and cowpea datasets. (a) Fitted $\Delta{H_{a,J_{max}}}$ versus fitted $\Delta{H_{a,V_{cmax}}}$. (b) Fitted $\Delta{H_{a,J_{max}}}$ versus fitted $\Delta{H_{a,\mathrm{TPU}}}$. Colored zones delineate bean and cowpea genotypes.
• Figure 5: Fitted $A_c/Q$, $A_j/Q$, and $A_p/Q$ curves, and measured $A/Q$ (black dot points) for light response type 1 (a) and 2(b). Curves in (a) and (b) are the same as curves from Fig. \ref{['fig:acil1']}j and \ref{['fig:acil2']}j, respectively.