---
title: 'SEAGAN: Graph Analysis of A–Ci Curves'
url: https://www.emergentmind.com/topics/seagan
type: topic
---

# SEAGAN: Graph Analysis of A–Ci Curves

SEAGAN denotes the **domain-Specific and Edge-Aware Graph Attention Network for Dynamic Plant Processes**, introduced for node-wise classification of biochemical limitation states along leaf-level photosynthetic \(A\)–\(C_i\) curves in C\(_3\) plants [2606.19623]. In this formulation, each curve is treated as a small graph whose nodes are measurement points and whose edges encode local relational structure in the \((C_i, A_\text{net})\) response space. The method addresses a specific bottleneck in \(A\)–\(C_i\) analysis: assigning each measured point to the active limitation regime—Rubisco limitation, electron-transport limitation, or triose-phosphate utilization limitation—before downstream parameter estimation under the Farquhar–von Caemmerer–Berry (FvCB) framework [2606.19623].

## 1. Biochemical setting and problem formulation

The target object is the **A–Ci curve**, which relates net CO\(_2\) assimilation rate \(A_\text{net}\) to intercellular CO\(_2\) concentration \(C_i\). Under the FvCB model for C\(_3\) photosynthesis, assimilation at a point is governed by the minimum of three candidate biochemical rates,
\[
A_\text{net} = \min\!\left\{A_c,\;A_j,\;A_p\right\},
\]
where \(A_c\) is the Rubisco-limited rate, \(A_j\) is the RuBP-regeneration or electron-transport-limited rate, and \(A_p\) is the triose-phosphate utilization-limited rate [2606.19623].

For a typical A–Ci curve, low \(C_i\) is associated with \(A_c\) limitation, intermediate \(C_i\) with \(A_j\) limitation, and high \(C_i\), under certain parameter regimes, with \(A_p\) limitation [2606.19623]. The classification of each point into one of these limitation states is operationally important because FvCB parameter estimation for \(V_\text{cmax}\), \(J\), TPU, \(g_m\), and \(R_d\) depends on fitting the correct portion of the curve under the correct regime. The paper states that mislabeling points near regime transitions causes biased or non-identifiable parameter estimates [2606.19623].

The difficulty of this labeling problem arises from several factors explicitly identified in the study: regime transitions are blurred in noisy data, gas-exchange measurements are noisy and subject to experimental artifacts, the FvCB model is switch-like because it uses a \(\min\{\cdot\}\) operator, and some curves weakly express one limitation state—especially TPU—so the data may contain little direct evidence for that regime [2606.19623]. Historically, practitioners have relied on heuristics or manual assignment, while automated fitting tools such as plantecophys, msuRACiFit, PhoTorch, and PhotoGEA attempt to infer limitation states and parameters jointly by numerical optimization. The paper’s motivation is that these approaches remain sensitive to regime misidentification and identifiability issues, and therefore a separate supervised learning formulation of limitation-state identification is warranted [2606.19623].

## 2. Graph representation of A–Ci curves

SEAGAN recasts each individual A–Ci curve as a graph \(G=(\mathcal{V},\mathcal{E})\), with each measurement point represented as a node. Curve lengths are small and variable, with \(m=8\)–15 nodes per curve after subsampling [2606.19623]. This graph construction is central rather than incidental: the paper’s stated key idea is to formulate limitation-state identification as a graph-based node classification problem so that local relational structure along the response curve can be exploited [2606.19623].

Each node \(j\) is assigned a four-dimensional feature vector
\[
\mathbf{x}_j =
\begin{bmatrix}
C_i(j)\\
A_\text{net}(j)\\
s_{Ac}(j)\\
s_{Aj}(j)
\end{bmatrix}
\in \mathbb{R}^{4},
\]
where \(C_i(j)\) and \(A_\text{net}(j)\) are the raw gas-exchange observations and \(s_{Ac}(j)\), \(s_{Aj}(j)\) are physiology-inspired auxiliary signals derived from FvCB structure using \(C_i\) in place of \(C_c\) [2606.19623]. The paper interprets these auxiliary quantities as effective capacities under Rubisco-like and electron-transport-like response factors. It further states that \(s_{Ac}\) tends to increase mainly within the Rubisco-limited region, whereas \(s_{Aj}\) increases through the \(A_c\) and \(A_j\) regions and then often flattens or declines as TPU limitation emerges [2606.19623]. Because noise prevents their use as hard thresholds, they are treated as informative features rather than deterministic indicators.

Two graph connectivity schemes are defined. The first is a **distance-based k-nearest-neighbor graph**, which is the configuration used by SEAGAN. For each node, Euclidean distance is computed in the observed response space \((C_i,A_\text{net})\), the \(k=4\) nearest neighbors are selected, directed edges are created to these neighbors, and reverse edges are added to make the graph bidirectional [2606.19623]. This produces a locally dense graph in which each node exchanges information with at least four neighbors close in the A–Ci plane; importantly, the points are deliberately shuffled so that node order does not trivially encode the curve trajectory [2606.19623].

The second scheme is the **auxiliary-signal-guided (ASG) graph**, used for comparison. In this construction, points are sorted by increasing \(C_i\), peaks are detected in the sorted auxiliary sequences, the curve is partitioned into three contiguous groups, nodes are fully connected within groups, and boundary edges are added between adjacent groups [2606.19623]. The study reports that SEAGAN ultimately uses the simpler kNN connectivity because it is performance-superior in the attention-based setting [2606.19623].

For every directed edge \((u,v)\), the edge attribute vector is defined as
\[
\mathbf{e}_{uv}=
\begin{bmatrix}
s_{Ac}(u)-s_{Ac}(v)\\
s_{Aj}(u)-s_{Aj}(v)
\end{bmatrix}
\in\mathbb{R}^{2}.
\]
These attributes encode pairwise changes in the auxiliary signals and therefore provide information about local slopes and relative positions in limitation space [2606.19623]. The reverse edge carries the negated attribute vector. This edge design is one of the paper’s principal domain-specific elements.

Ground-truth node labels are assigned directly from the synthetic FvCB generative model: class 0 corresponds to \(A_c\)-limited, class 1 to \(A_j\)-limited, and class 2 to \(A_p\)-limited, with the label at each point determined by the index of the minimum among \(A_c\), \(A_j\), and \(A_p\) [2606.19623].

## 3. Architecture and domain-specific design

SEAGAN is identified in the paper as the best-performing configuration among the compared models. Architecturally, it is a **Graph Attention Network (GAT)** operating on the kNN graph with edge attributes and trained using weighted cross-entropy loss; in the code naming reported by the paper, it corresponds to **GAT-kNN-\(\mathcal{L}_\mathrm{WCE}\)** [2606.19623].

The base message-passing mechanism follows the standard multi-head GAT formulation. Let \(\mathbf{h}_j^{(\ell)}\) denote the hidden representation of node \(j\) at layer \(\ell\), and let \(K=5\) be the number of attention heads. For intermediate layers,
\[
\mathbf{h}_j^{(\ell+1)} =
\mathop{\Vert}_{k=1}^{K}
\sigma\left(
\sum_{q\in\mathcal{N}(j)}
\alpha_{jq}^{(\ell,k)}
\,\mathbf{W}^{(\ell,k)}\mathbf{h}_q^{(\ell)}
\right),
\]
while in the final GAT layer the heads are averaged rather than concatenated [2606.19623]. The nonlinear activation is tanh, batch normalization is applied after graph layers, and dropout with \(p=0.2\) is used for regularization [2606.19623]. The number of layers \(L\) is not explicitly stated, but the paper notes that at least two layers are used because it distinguishes intermediate and final layers [2606.19623].

The attention mechanism is **edge-aware**. Although the paper does not provide an explicit symbolic expression for the attention coefficients \(\alpha_{jq}^{(\ell,k)}\), it states that the attention kernel receives the transformed target-node representation, the transformed neighbor representation, and the edge attribute \(\mathbf{e}_{jq}\) [2606.19623]. In consequence, the model can modulate neighborhood aggregation using both node states and pairwise physiological differences. The study interprets this as allowing the network to emphasize edges whose auxiliary-signal differences support a particular limitation regime, especially around ambiguous transition points [2606.19623].

After the final GAT layer, node embeddings are mapped to class logits through a linear classifier,
\[
\mathbf{Z}=\mathbf{H}^{(L)}\mathbf{W}_\mathrm{out}+\mathbf{b}_\mathrm{out},
\]
and prediction for node \(j\) is obtained by
\[
\hat{y}_j=\arg\max_{c\in\{0,1,2\}} Z_{j,c}.
\]
Softmax over \(\mathbf{Z}\) yields class probabilities used for losses and interpretability analyses [2606.19623].

The paper explicitly locates the distinctiveness of SEAGAN not in a novel generic GAT backbone, but in its **domain-specific representation design**. Three elements are emphasized: physiology-aware node features through inclusion of \(s_{Ac}\) and \(s_{Aj}\), physiology-aware edge attributes defined from differences in those auxiliary signals, and physiology-aware connectivity built in the \((C_i,A_\text{net})\) plane [2606.19623]. The authors describe these design choices as what makes the GNN process-aware.

## 4. Training protocol and synthetic benchmark construction

The training objective used by SEAGAN is a **weighted cross-entropy loss** to compensate for imbalance among the three classes. The standard multi-class cross-entropy is modified by class weights \(w_c\), computed from training class frequencies and inversely related to class prevalence, yielding
\[
\mathcal{L}_{\mathrm{WCE}}
=
-
\sum_{j=1}^{N_T}\sum_{c=1}^{C}
w_c\,y_{jc}\log(\hat{y}_{jc}),
\]
with \(C=3\) classes [2606.19623]. The study also evaluates a weighted focal loss, but SEAGAN specifically uses \(\mathcal{L}_\mathrm{WCE}\) [2606.19623].

The dataset consists of **10,000 synthetic curves**, each generated on a dense grid of 100 \(C_i\) points spanning 20 to 1000 ppm and then subsampled to 8–15 points to mimic real experiments [2606.19623]. The subsampled points are shuffled so that increasing \(C_i\) order is not preserved [2606.19623]. Parameter values at \(25^\circ\)C are sampled by Latin Hypercube Sampling over the ranges stated in the paper: \(V_{\text{cmax},25}\) from 20 to 150 \(\mu\)mol m\(^{-2}\) s\(^{-1}\), \(\mathrm{TPU}_{25}\) from 4 to 25 \(\mu\)mol m\(^{-2}\) s\(^{-1}\), \(R_{\text{d},25}\) from 0.5 to 5 \(\mu\)mol m\(^{-2}\) s\(^{-1}\), \(g_{\text{m},25}\) from 0.05 to 0.8 mol m\(^{-2}\) s\(^{-1}\), \(J_{\text{max},25}/V_{\text{cmax},25}\) from 1.5 to 2.5, and leaf temperature \(T_\text{leaf}\) from 25 to 35\(^\circ\)C [2606.19623]. Environmental conditions are fixed at \(P_\text{atm}=101.325\) kPa and O\(_2\) \(=210\) mbar [2606.19623]. Temperature response functions for \(V_\text{cmax}, J, \mathrm{TPU}, R_d, g_m, K_c, K_o,\Gamma^\ast\) are taken from Sharkey et al. and Bernacchi et al. [2606.19623].

To mimic measurement error, Gaussian noise is added to \(A_\text{net}\). The paper reports a mean curve-wise standard deviation of added noise of approximately 1.5 \(\mu\)mol m\(^{-2}\) s\(^{-1}\), corresponding to a normalized noise magnitude of \(0.10\), or 10% of each curve’s original \(A_\text{net}\) variability [2606.19623]. The authors characterize this level as introducing noticeable variability without dominating the signal.

Training, validation, and test splits are made at the **curve level** as 60%, 20%, and 20%, respectively, preventing leakage across splits [2606.19623]. Node features and edge attributes are z-score standardized using the training-set mean and standard deviation. Optimization uses Adam with learning rate 0.001 and weight decay 0.0001; batches contain 128 curves; non-U-Net models are trained for 800–1000 epochs; and the best checkpoint is selected by validation macro-F1 rather than by classical early stopping [2606.19623]. Each configuration is trained with 30 random initializations, and the resulting performance distributions are compared statistically using ANOVA and Tukey–HSD [2606.19623].

A methodological consequence of this setup is that SEAGAN is evaluated under conditions where ground-truth limitation states are fully known from the generative model. This suggests that the reported results isolate the regime-classification problem itself rather than confounding it with manual labeling inconsistencies.

## 5. Comparative evaluation and empirical performance

The study compares SEAGAN against feature-based baselines, graph baselines, and an automated fitting-based benchmark [2606.19623]. The feature-based models are Random Forest, Support Vector Machine with polynomial or RBF kernels, XGBoost, and a feed-forward neural network. These models take flattened vectors of length \(4m\) as inputs and produce \(3m\) outputs, with a separate model trained for each curve length \(m\in[8,15]\) [2606.19623]. The graph baselines include **GCN-kNN** and Graph U-Net variants based on GCN or GAT with kNN or ASG connectivity and several loss functions [2606.19623]. The fitting-based benchmark is **PhoTorch**, which is run on each noisy synthetic curve to fit FvCB parameters and then derive limitation-state labels from the fitted \(A_c\), \(A_j\), and \(A_p\) values [2606.19623].

The held-out test-set metrics reported in the paper are summarized below [2606.19623].

| Model | F1 | Accuracy |
|---|---:|---:|
| PhoTorch benchmark | 0.735 | 0.751 |
| Best feature-based NN baseline | 0.713 | 0.766 |
| Baseline GCN-kNN \((\mathcal{L}_\mathrm{CE})\) | 0.722 | 0.778 |
| Best GCN-U-Net-ASG \((\mathcal{L}_\mathrm{CE})\) | 0.817 | 0.851 |
| SEAGAN: GAT-kNN-\(\mathcal{L}_\mathrm{WCE}\) | 0.857 | 0.882 |

For SEAGAN specifically, the paper reports **F1 = 0.857**, **recall = 0.861**, **precision = 0.853**, and **accuracy = 0.882** [2606.19623]. Closely related GAT-kNN variants using unweighted cross-entropy or weighted focal loss perform similarly, with F1 values of 0.854 and 0.855, respectively [2606.19623]. The Tukey–HSD analysis over 12 GNN models and 30 runs places the GAT-kNN variants in the top statistical groups, and the paper characterizes GAT-kNN-\(\mathcal{L}_\mathrm{WCE}\) as the best compromise between highest mean F1 and recall and robustness, with fewer low-performing outliers [2606.19623].

The paper also reports qualitative behavior near transition regions. Feature-based baselines often place the \(A_c\to A_j\) transition too late and the \(A_j\to A_p\) transition too early, and they can produce isolated mislabels inside otherwise homogeneous regions [2606.19623]. GCN-kNN improves transition handling but still misclassifies some boundary points. In a representative example, PhoTorch shifts the \(A_c\to A_j\) transition earlier than ground truth, whereas SEAGAN correctly recovers the transition node and assigns \(p(A_j)=0.934\) versus \(p(A_c)=0.043\) for that node [2606.19623]. Even so, the paper notes that some \(A_j\to A_p\) transitions remain difficult for all models, including SEAGAN and PhoTorch, especially when TPU limitation is weakly expressed [2606.19623].

Several comparative conclusions are drawn directly from the reported experiments. First, graph structure itself adds value, because GCN-kNN exceeds the SVM, XGBoost, and feed-forward NN baselines in F1 [2606.19623]. Second, attention outperforms fixed graph convolution, as GAT variants substantially exceed GCN and GCN-U-Net variants [2606.19623]. Third, kNN connectivity outperforms ASG connectivity in GAT models by several points in F1 and accuracy, with statistically significant differences [2606.19623]. Fourth, hierarchical pooling improves GCN but does not improve plain GAT-kNN, which the paper plausibly attributes to the small size of the graphs and the local nature of the discriminative information [2606.19623].

## 6. Interpretability, physiological significance, and limitations

Interpretability analysis is performed with **GNNExplainer**, which is used to identify influential edges for a given node prediction and to perform edge ablation by perturbing those edges and measuring probability changes, particularly \(\Delta p_{A_j}-\Delta p_{A_c}\) around transition nodes [2606.19623]. For the GCN baseline, the important edges associated with a transition node tend to support the wrong class and are less focused on physiologically relevant neighbors [2606.19623]. For GAT-kNN, the important edges connect the ambiguous node to neighbors on both sides of the \(A_c\to A_j\) boundary, and ablating those edges reduces the probability of the correct class, indicating that they positively support correct classification [2606.19623]. GAT-ASG also improves over GCN but relies more on self or within-group connections and provides less robust cross-regime information than GAT-kNN [2606.19623].

The paper interprets these findings physiologically. SEAGAN appears to learn that transitions are best resolved by comparing ambiguous points with nearby points across regimes, and the kNN graph supplies those cross-boundary neighbors while the edge attributes describe differences in the auxiliary signals \(s_{Ac}\) and \(s_{Aj}\) [2606.19623]. The resulting predictions better preserve contiguous regions corresponding to each limitation state, which the authors connect to the physical expectation that biochemical regimes change smoothly along the response trajectory rather than through isolated jumps [2606.19623].

An important misconception addressed by the study is that SEAGAN is not a replacement for biochemical parameter estimation. The authors explicitly frame it as a regime-classification model rather than a direct estimator of \(V_\text{cmax}\), \(J\), TPU, \(g_m\), or \(R_d\) [2606.19623]. A plausible implication is that its main value lies in decoupling classification from fitting, thereby providing a cleaner upstream signal for downstream parameter inference.

The paper also enumerates several limitations. Validation is conducted only on synthetic curves, so generalization to real experimental A–Ci data, with different noise patterns, species, measurement protocols, or systematic biases, remains untested [2606.19623]. The graphs are small, and no scalability experiments are reported beyond the 8–15 node regime [2606.19623]. Transition ambiguity persists, especially for weakly expressed TPU limitation [2606.19623]. Graph construction assumptions may also matter: kNN in \((C_i,A_\text{net})\) space may not be optimal under heavy noise or unusual sampling, and ASG connectivity depends on peak detection in auxiliary signals, which may be noise-sensitive [2606.19623].

Within those bounds, the authors position SEAGAN as a pre-fitting component for automated photosynthesis analysis pipelines. They suggest that fitting tools such as PhoTorch, PhotoGEA, and plantecophys could use SEAGAN predictions to restrict fitting to regime-appropriate points, constrain transition locations, or initialize or regularize change-point models [2606.19623]. The paper further proposes that the general methodology—graph representation of response trajectories with process-aware features and edge attributes—may extend to other dynamic plant processes, including stomatal responses to VPD, light-response curves, and chlorophyll fluorescence kinetics [2606.19623]. This suggests a broader research program in which small graphs encode local structure in physiological trajectories and edge-aware attention is used to resolve latent process states.

Source: https://www.emergentmind.com/topics/seagan