From lipid dynamics to precision predictions: A new approach methodology for precision modeling of phosphoinositide signaling.

Code ↔ Paper

11 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 11 matches · 3 of them tie a paragraph to a whole file, not to given lines: weak matches, whose lines are not tinted

[1] § Methods › Phosphoinositide signaling model structure ↔ Codes_For_Figures/Figure_4/Global_Sensitivity_Analysis.m, lines 1–42 · score 0.81 · plasma membrane surface, cytoplasmic volume, converts molecules, conversion factor, biosensor, Phosphoinositide
[2] § Methods › Phosphoinositide signaling model structure ↔ Codes_For_Figures/Figure_2/Figure_2_B.m, lines 1–55 · score 0.81 · plasma membrane surface, cytoplasmic volume, converts molecules, conversion factor, biosensor, signaling
[3] § Methods › Neural network parameter estimation ↔ Codes_For_Figures/Figure_6/Figure_6_AB.m, the whole file · a weak match · score 0.78 · neural network, network predicted, kPLC, kPIP5K, kPI4K, kdeg
[4] § Methods › Neural network parameter estimation ↔ Codes_For_Figures/Figure_6/Colab/PI_Cycle_v6.ipynb, lines 279–289 · score 0.78 · learning rate scheduling, weight decay, PyTorch, Adam, patience, loss
[5] § Results ↔ Codes_For_Figures/Figure_6/Figure_6_AB.m, the whole file · a weak match · score 0.73 · neural network predicted, parameter fingerprints, kPIP5K, kPI4K, k4P, neuroblastoma
[6] § Methods › Neural network parameter estimation ↔ Codes_For_Figures/Figure_6/Colab/PI_Cycle_v6.ipynb, lines 140–204 · score 0.69 · Latin Hypercube, stimulus durations, kinetic parameters, uniformly, channel, 40 %
[7] § Methods › Global sensitivity analysis ↔ Codes_For_Figures/Figure_4/Global_Sensitivity_Analysis.m, lines 44–101 · score 0.63 · Latin Hypercube Sampling, global sensitivity, LHS
[8] § Methods › Neural network parameter estimation ↔ Codes_For_Figures/Figure_6/Colab/PI_Cycle_v6.ipynb, lines 140–204 · score 0.59 · stimulus duration, steady state, enforce, seeds, prediction, baseline
[9] § Methods › Phosphoinositide signaling model structure ↔ Codes_For_Figures/Figure_2/Isolated_SCG_Neurons.m, lines 160–200 · score 0.53 · state variables, LIBRAvIII, cytoplasmic, membrane, bound, molecules
[10] § Methods › Nelder-Mead parameter optimization of the phosphoinositide signaling model ↔ Codes_For_Figures/Figure_2/Figure_2_A_Optimization/main_optimization.m, the whole file · a weak match · score 0.52 · squared errors, sum, SSE, global, optimization, model
[11] § Methods › Neural network parameter estimation ↔ Codes_For_Figures/Figure_6/Colab/PI_Cycle_v6.ipynb, lines 224–268 · score 0.52 · hidden, kernel, layers, LSTM, batch, max

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The paper is loaded by your browser, with JavaScript: read it at doi.org or on Europe PMC.

The authors' code

Jupyter notebook · 767 lines · 32 KB · GPL-3.0 · 4 matches

# %% [markdown]
# # PI Cycle Parameter Estimation - v4
#
# **Key change from v3: ratio fed as input, not predicted**
#
# The PI/PIP ratio is known from biology for each cell type:
# - SCG: 50.0
# - tsA: 35.0
# - Neuroblastoma: 22.42
# - Hippocampal: 5.38
#
# Instead of asking the network to guess the ratio from normalized data
# (impossible after normalization removes absolute pool size), we feed it
# as a 5th input channel. The network then predicts only 5 parameters:
# k_PI4K, k_PIP5K, k_5P, k_plc, k_deg.
#
# k_4P is always derived as: k_4P = known_ratio * predicted_k_PI4K
#
# **Run cells in order. Do not skip.**
# %%
# -------------------------------------------------------
# SET RUN SEED HERE BEFORE HITTING RUN ALL
# Run 1: RUN_SEED = 42
# Run 2: RUN_SEED = 123
# Run 3: RUN_SEED = 456
# -------------------------------------------------------
RUN_SEED = 456
# -------------------------------------------------------
# Uncomment below ONLY to delete checkpoint for this run
# import os
# for f in ['best_model.pth', 'training_checkpoint.pt']:
# path = f'/content/drive/MyDrive/PI_Cycle_checkpoints/run_{RUN_SEED}/{f}'
# if os.path.exists(path):
# os.remove(path)
# print(f'Deleted {f}')
print(f'RUN_SEED = {RUN_SEED}')
# %%
!pip install -q torch numpy scipy matplotlib scikit-learn tqdm
print('Packages ready!')
# %%
import numpy as np
import matplotlib.pyplot as plt
from scipy.integrate import odeint
from scipy.stats import qmc
from sklearn.model_selection import train_test_split
import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
from tqdm import tqdm
import time
import os
from google.colab import drive
drive.mount('/content/drive')
OUTPUT_DIR = f'/content/drive/MyDrive/PI_Cycle_outputs/run_{RUN_SEED}'
CHECKPOINT_DIR = f'/content/drive/MyDrive/PI_Cycle_checkpoints/run_{RUN_SEED}'
os.makedirs(OUTPUT_DIR, exist_ok=True)
os.makedirs(CHECKPOINT_DIR, exist_ok=True)
print('Imports done, Drive mounted.')
# %%
# -------------------------------------------------------
# TOGGLE HERE
# True = quick test: 10k samples, 30 epochs (~10 min)
# False = full run: 50k samples, 50 epochs (~60 min)
# -------------------------------------------------------
TEST_MODE = False
# -------------------------------------------------------
# -------------------------------------------------------
N_SAMPLES = 10000 if TEST_MODE else 50000
N_EPOCHS = 30 if TEST_MODE else 50
PATIENCE = 10
CELL_RATIOS = {
'SCG': 50.00,
'tsA': 35.00,
'Neuroblastoma': 22.42,
'Hippocampal': 5.38,
}
RATIO_MIN = 3.0
RATIO_MAX = 50.0
def normalize_ratio(ratio):
return (ratio - RATIO_MIN) / (RATIO_MAX - RATIO_MIN)
print(f'RUN_SEED: {RUN_SEED}')
print(f'Mode: {"TEST" if TEST_MODE else "FULL"} | Samples: {N_SAMPLES:,} | Epochs: {N_EPOCHS}')
# %%
def pi_cycle_odes(y, t, params, PI, cf,
stim_start=80, stim_end=100, PLC_amp=0.7):
PIP, PIP2, IP3_free, LIBRA_free, IP3_LIBRA = y
k_PI4K = params['k_PI4K']
k_4P = params['k_4P']
k_PIP5K = params['k_PIP5K']
k_5P = params['k_5P']
k_plc = params['k_plc']
k_deg = params['k_deg']
PLC = PLC_amp if (stim_start <= t <= stim_end) else 0.0
spd = 1.0
KD = 0.5
fp = 1.0
dPIP = fp*k_PI4K*PI - k_4P*PIP - fp*k_PIP5K*PIP + fp*k_5P*PIP2
dPIP2 = fp*k_PIP5K*PIP - fp*k_5P*PIP2 - k_plc*fp*PIP2*PLC
dIP3 = cf*fp*k_plc*PIP2*PLC - k_deg*IP3_free - spd*IP3_free*LIBRA_free + spd*KD*IP3_LIBRA
dLIBRA = -spd*IP3_free*LIBRA_free + spd*KD*IP3_LIBRA
dIP3L = spd*LIBRA_free*IP3_free - spd*KD*IP3_LIBRA
return [dPIP, dPIP2, dIP3, dLIBRA, dIP3L]
def steady_state_y0(k_PI4K, PI_PIP_ratio, k_PIP5K, k_5P, PI):
"""Exact pre-stimulus steady state. IP3=0 before stimulus."""
PIP_ss = PI / PI_PIP_ratio
PIP2_ss = (k_PIP5K / k_5P) * PIP_ss
return [PIP_ss, PIP2_ss, 0.0, 6.0, 0.0]
print('ODE system defined.')
# %%
baseline = {
'k_PI4K': 0.00079,
'PI_PIP_ratio': 50.0,
'k_PIP5K': 0.01624,
'k_5P': 0.02172,
'k_plc': 0.242,
'k_deg': 0.07
}
print('Baseline parameters set.')
# %%
def generate_data(n_samples=50000, seed=42):
np.random.seed(seed)
print(f'Generating {n_samples:,} samples...')
PI = 226975
SA = 4100
Vol = 6644
cf = (SA / Vol) * (1 / 602)
# 5 kinetic parameters to predict (ratio removed from labels)
ranges = {
'k_PI4K': (0.6 * baseline['k_PI4K'], 1.4 * baseline['k_PI4K']),
'k_PIP5K': (0.4 * baseline['k_PIP5K'], 1.6 * baseline['k_PIP5K']),
'k_5P': (0.4 * baseline['k_5P'], 1.6 * baseline['k_5P']),
'k_plc': (0.4 * baseline['k_plc'], 1.6 * baseline['k_plc']),
'k_deg': (0.6 * baseline['k_deg'], 1.4 * baseline['k_deg']),
}
stim_durations = np.random.choice([20, 50, 60], size=n_samples)
# Sample ratio separately -- uniform over full range
ratios = np.random.uniform(RATIO_MIN, RATIO_MAX, size=n_samples)
# LHS for the 5 kinetic parameters
sampler = qmc.LatinHypercube(d=5, seed=seed)
raw = sampler.random(n=n_samples)
p = np.zeros_like(raw)
for i, (lo, hi) in enumerate(ranges.values()):
p[:, i] = raw[:, i] * (hi - lo) + lo
X, y = [], []
for i in tqdm(range(n_samples)):
k_PI4K, k_PIP5K, k_5P, k_plc, k_deg = p[i]
ratio = ratios[i]
k_4P = ratio * k_PI4K # constraint enforced
y0 = steady_state_y0(k_PI4K, ratio, k_PIP5K, k_5P, PI)
pdict = {'k_PI4K': k_PI4K, 'k_4P': k_4P, 'k_PIP5K': k_PIP5K,
'k_5P': k_5P, 'k_plc': k_plc, 'k_deg': k_deg}
sd = stim_durations[i]
t = np.linspace(0, 300, 100)
try:
sol = odeint(pi_cycle_odes, y0, t,
args=(pdict, PI, cf, 80, 80+sd, 0.7),
rtol=1e-8, atol=1e-10)
if not np.isfinite(sol).all():
continue
PIP = sol[:, 0] / sol[0, 0]
PIP2 = sol[:, 1] / sol[0, 1]
mx = sol[:, 4].max()
IP3 = sol[:, 4] / mx if mx > 0 else sol[:, 4]
stim = np.full(100, sd / 100.0)
# 5th channel: normalized ratio (same value across all 100 time points)
ratio_ch = np.full(100, normalize_ratio(ratio))
X.append(np.stack([PIP, PIP2, IP3, stim, ratio_ch], axis=1))
# Labels: 5 kinetic params only (ratio is an input, not a target)
y.append([k_PI4K, k_PIP5K, k_5P, k_plc, k_deg])
except Exception:
continue
print(f'Valid samples: {len(X):,} / {n_samples:,}')
return np.array(X, dtype=np.float32), np.array(y, dtype=np.float32)
print('Data generation function ready.')
# %%
X_train, y_train = generate_data(n_samples=N_SAMPLES, seed=RUN_SEED)
print(f'X: {X_train.shape} # (n, 100, 5)')
print(f'y: {y_train.shape} # (n, 5)')
# %%
fig, axes = plt.subplots(1, 3, figsize=(15, 4))
t = np.linspace(0, 300, 100)
for j, name in enumerate(['PIP', 'PIP2', 'IP3']):
for i in np.random.choice(len(X_train), 100, replace=False):
axes[j].plot(t, X_train[i, :, j], alpha=0.05, color='blue')
axes[j].set_title(name)
axes[j].set_xlabel('Time (s)')
axes[j].grid(alpha=0.3)
plt.tight_layout()
plt.savefig(f'{OUTPUT_DIR}/training_preview.png', dpi=150)
plt.show()
# %%
class PIDataset(Dataset):
def __init__(self, X, y):
self.X = torch.FloatTensor(X)
self.y_log = torch.log10(torch.FloatTensor(y) + 1e-10)
def __len__(self):
return len(self.X)
def __getitem__(self, idx):
return self.X[idx], self.y_log[idx]
class PINet(nn.Module):
def __init__(self, input_dim=5, hidden_dim=128, num_layers=2, output_dim=5):
super().__init__()
self.conv1 = nn.Conv1d(input_dim, 64, kernel_size=5, padding=2)
self.conv2 = nn.Conv1d(64, 128, kernel_size=5, padding=2)
self.pool = nn.MaxPool1d(2)
self.lstm = nn.LSTM(128, hidden_dim, num_layers,
batch_first=True, dropout=0.2)
self.fc1 = nn.Linear(hidden_dim, 128)
self.fc2 = nn.Linear(128, 64)
self.fc3 = nn.Linear(64, output_dim)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(0.3)
self.bn1 = nn.BatchNorm1d(64)
self.bn2 = nn.BatchNorm1d(128)
def forward(self, x):
x = x.permute(0, 2, 1)
x = self.relu(self.bn1(self.conv1(x)))
x = self.pool(x)
x = self.relu(self.bn2(self.conv2(x)))
x = self.pool(x)
x = x.permute(0, 2, 1)
x, _ = self.lstm(x)
x = x[:, -1, :]
x = self.relu(self.fc1(x))
x = self.dropout(x)
x = self.relu(self.fc2(x))
x = self.dropout(x)
return self.fc3(x)
print('Dataset and model defined.')
print('Input channels: PIP, PIP2, IP3, stim_duration, ratio (5 total)')
print('Output params: k_PI4K, k_PIP5K, k_5P, k_plc, k_deg (5 total)')
# %%
X_tr, X_val, y_tr, y_val = train_test_split(
X_train, y_train, test_size=0.1, random_state=RUN_SEED)
print(f'Train: {len(X_tr):,} Val: {len(X_val):,}')
train_loader = DataLoader(PIDataset(X_tr, y_tr), batch_size=64, shuffle=True)
val_loader = DataLoader(PIDataset(X_val, y_val), batch_size=64, shuffle=False)
print('Loaders ready.')
# %%
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(f'Device: {device}')
model = PINet(input_dim=5, output_dim=5).to(device)
criterion = nn.MSELoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.001, weight_decay=1e-5)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
optimizer, mode='min', factor=0.5, patience=5)
print(f'Parameters: {sum(p.numel() for p in model.parameters()):,}')
# %%
def train_epoch(model, loader, criterion, optimizer, device):
model.train()
total_loss, valid_batches = 0.0, 0
for X_batch, y_batch in loader:
X_batch, y_batch = X_batch.to(device), y_batch.to(device)
if not torch.isfinite(X_batch).all(): continue
if not torch.isfinite(y_batch).all(): continue
optimizer.zero_grad()
loss = criterion(model(X_batch), y_batch)
if not torch.isfinite(loss): continue
loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)
optimizer.step()
total_loss += loss.item()
valid_batches += 1
return total_loss / valid_batches if valid_batches > 0 else float('nan')
def validate(model, loader, criterion, device):
model.eval()
total_loss, valid_batches = 0.0, 0
with torch.no_grad():
for X_batch, y_batch in loader:
X_batch, y_batch = X_batch.to(device), y_batch.to(device)
if not torch.isfinite(X_batch).all(): continue
if not torch.isfinite(y_batch).all(): continue
loss = criterion(model(X_batch), y_batch)
if not torch.isfinite(loss): continue
total_loss += loss.item()
valid_batches += 1
return total_loss / valid_batches if valid_batches > 0 else float('nan')
print('Training functions ready.')
# %%
checkpoint_path = f'{CHECKPOINT_DIR}/training_checkpoint.pt'
start_epoch = 0
best_val_loss = float('inf')
train_losses = []
val_losses = []
patience_counter = 0
if os.path.exists(checkpoint_path):
print('Found checkpoint, resuming...')
ckpt = torch.load(checkpoint_path)
model.load_state_dict(ckpt['model'])
optimizer.load_state_dict(ckpt['optimizer'])
scheduler.load_state_dict(ckpt['scheduler'])
start_epoch = ckpt['epoch'] + 1
best_val_loss = ckpt['best_val_loss']
train_losses = ckpt['train_losses']
val_losses = ckpt['val_losses']
patience_counter = ckpt['patience_counter']
print(f'Resumed from epoch {start_epoch}')
else:
print('Starting fresh training')
print(f'Training for up to {N_EPOCHS} epochs...\n')
start_time = time.time()
for epoch in range(start_epoch, N_EPOCHS):
train_loss = train_epoch(model, train_loader, criterion, optimizer, device)
val_loss = validate(model, val_loader, criterion, device)
train_losses.append(train_loss)
val_losses.append(val_loss)
if np.isfinite(val_loss) and val_loss < best_val_loss:
best_val_loss = val_loss
torch.save(model.state_dict(), f'{CHECKPOINT_DIR}/best_model.pth')
patience_counter = 0
else:
patience_counter += 1
torch.save({
'epoch': epoch, 'model': model.state_dict(),
'optimizer': optimizer.state_dict(), 'scheduler': scheduler.state_dict(),
'best_val_loss': best_val_loss, 'train_losses': train_losses,
'val_losses': val_losses, 'patience_counter': patience_counter
}, checkpoint_path)
if (epoch + 1) % 10 == 0:
lr = optimizer.param_groups[0]['lr']
print(f'Epoch {epoch+1}/{N_EPOCHS} | '
f'Train: {train_loss:.6f} | Val: {val_loss:.6f} | '
f'Best: {best_val_loss:.6f} | LR: {lr:.6f}')
if patience_counter >= PATIENCE:
print(f'Early stopping at epoch {epoch+1}')
break
elapsed = time.time() - start_time
print(f'\nDone in {elapsed/60:.1f} min | Best val loss: {best_val_loss:.6f}')
# %%
plt.figure(figsize=(8, 4))
plt.plot(train_losses, label='Train')
plt.plot(val_losses, label='Val')
plt.xlabel('Epoch')
plt.ylabel('MSE Loss')
plt.title('Training curve')
plt.legend()
plt.grid(alpha=0.3)
plt.tight_layout()
plt.savefig(f'{OUTPUT_DIR}/loss_curve.png', dpi=150)
plt.show()
# %%
experimental_data = {
'SCG': {
'time': np.arange(0, 264, 4),
'PIP': np.array([1.000228945, 1.019145079, 1.01847486, 1.013001365, 1.011373365, 1.00949151, 1.007427164, 1.013568162, 1.007596261, 1.007161807, 1.006393468, 1.005397447, 1.005972331, 1.004322455, 1.00322294, 1.001841359, 0.999476243, 0.994473405, 0.985835205, 0.978447035, 0.95768421, 0.92601231, 0.883451217, 0.840965043, 0.794204451, 0.768181235, 0.746076956, 0.740754728, 0.735236155, 0.738837119, 0.738899465, 0.738881756, 0.735339572, 0.739057224, 0.748468351, 0.755294914, 0.762498274, 0.769645878, 0.788341545, 0.796131065, 0.806043529, 0.816959776, 0.831344526, 0.837966439, 0.856660122, 0.866097364, 0.873055981, 0.888776333, 0.897794697, 0.898283764, 0.908599388, 0.912732096, 0.918565414, 0.927323747, 0.931540278, 0.930044926, 0.935394321, 0.941467019, 0.947825275, 0.950261638, 0.95221063, 0.953447485, 0.955094919, 0.962056348, 0.961344415, 0.964188999]),
'PIP2': np.array([1.00000084, 0.994329381, 1.003684332, 1.01238027, 1.001370793, 1.014275208, 1.019068623, 1.025286393, 1.021376017, 1.033808349, 1.033765832, 1.044470898, 1.057564312, 1.066161852, 1.054367241, 1.054415043, 1.03169049, 1.03743873, 1.050858443, 1.04603695, 1.029628345, 0.923978784, 0.48679318, 0.276771394, 0.137688215, 0.087431413, 0.050632959, 0.049037994, 0.076933626, 0.1173743, 0.167048938, 0.223215209, 0.273250572, 0.329744052, 0.370604954, 0.432028621, 0.469869382, 0.488020552, 0.556228878, 0.593654941, 0.640071437, 0.663253074, 0.69453996, 0.716814579, 0.729178755, 0.760902464, 0.772180962, 0.80298907, 0.816698667, 0.835885027, 0.855669997, 0.881415341, 0.867530616, 0.881964245, 0.89338665, 0.867959341, 0.873557793, 0.882656851, 0.888458335, 0.910417653, 0.885229072, 0.902716717, 0.889080619, 0.884813434, 0.899757591, 0.8871869]),
'IP3': np.array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.187654148, 0.488781155, 0.797607295, 0.889403036, 0.965747014, 1, 0.95862008, 0.965747014, 0.907691096, 0.918967978, 0.890175961, 0.757785213, 0.79063917, 0.724851376, 0.647930747, 0.692170804, 0.592637375, 0.513760867, 0.544461898, 0.635768453, 0.4959248, 0.454274312, 0.4162112, 0.424496157, 0.436885106, 0.371886584, 0.354061644, 0.360037911, 0.343798328, 0.314046331, 0.310208367, 0.311755623, 0.208950662, 0.224153337, 0.267904163, 0.254036531, 0.223629547, 0.199424576, 0.229936645, 0.156234824, 0.152500174, 0.12721983, 0.050564457, 0.120429132, 0.120429132]),
'stimulus_duration': 20,
'PI_PIP_ratio': 50.00
},
'Hippocampal': {
'time': np.arange(0, 276, 4),
'PIP': np.array([1.02, 1.02, 1.02, 1.02, 1.03, 1.03, 1.02, 1.02, 1.02, 1.02, 1.02, 1.01, 1.01, 1.01, 1.00, 1.00, 1.01, 1.00, 1.00, 1.00, 1.00, 1.00, 0.99, 0.99, 0.99, 0.98, 0.98, 0.99, 0.99, 0.99, 0.97, 0.98, 0.97, 0.97, 0.96, 0.96, 0.95, 0.95, 0.95, 0.93, 0.94, 0.92, 0.93, 0.93, 0.93, 0.93, 0.92, 0.91, 0.92, 0.92, 0.91, 0.90, 0.92, 0.91, 0.91, 0.90, 0.90, 0.90, 0.91, 0.89, 0.90, 0.88, 0.89, 0.90, 0.89, 0.90, 0.88, 0.88, 0.88]),
'PIP2': np.array([0.914, 0.914, 0.914, 0.914, 0.965, 0.963, 0.967, 0.958, 1.002, 0.965, 0.963, 0.967, 0.958, 1.002, 0.974, 1.032, 1.011, 1.002, 1.034, 1.053, 1.030, 0.958, 0.737, 0.410, 0.201, 0.141, 0.008, 0.055, 0.039, 0.032, -0.008, 0.041, 0.133, 0.184, 0.147, 0.147, 0.168, 0.187, 0.235, 0.307, 0.377, 0.451, 0.507, 0.574, 0.627, 0.664, 0.711, 0.710, 0.729, 0.752, 0.764, 0.768, 0.773, 0.780, 0.803, 0.768, 0.770, 0.801, 0.768, 0.768, 0.773, 0.888, 0.854, 0.791, 0.798, 0.733, 0.721, 0.751, 0.742]),
'IP3': np.array([0.147059, 0.147059, 0.147059, 0.147059, 0.117647, 0.029412, 0, 0.029412, 0.073529, 0.044118, 0.073529, 0.029412, 0.029412, 0.058824, 0.073529, 0.102941, 0.132353, 0.058824, 0.014706, 0.161765, 0.323529, 0.426471, 0.661765, 0.720588, 0.808824, 0.897059, 0.926471, 0.897059, 0.941176, 1, 0.970588, 0.985294, 0.955882, 0.985294, 1, 0.970588, 0.955882, 0.882353, 0.838235, 0.867647, 0.779412, 0.632353, 0.632353, 0.544118, 0.455882, 0.205882, 0.235294, 0.235294, 0.132353, 0.088235, 0.073529, 0.235294, 0.25, 0.220588, 0.161765, 0.058824, 0.102941, 0.132353, 0.058824, 0.132353, 0.161765, 0.102941, 0.088235, 0.102941, 0.132353, 0.102941, 0.147059, 0.205882, 0.25]),
'stimulus_duration': 50,
'PI_PIP_ratio': 5.38
},
'tsA': {
'time': np.arange(0, 300, 4),
'PIP': np.array([0.946, 0.962, 0.983, 0.962, 0.965, 0.988, 0.958, 0.974, 0.957, 0.972, 0.971, 1, 0.979, 0.997, 0.989, 0.994, 0.993, 0.966, 0.956, 0.903, 0.873, 0.814, 0.718, 0.683, 0.662, 0.654, 0.643, 0.657, 0.606, 0.612, 0.6, 0.602, 0.604, 0.6, 0.606, 0.605, 0.614, 0.62, 0.635, 0.641, 0.631, 0.635, 0.64, 0.656, 0.628, 0.64, 0.661, 0.661, 0.676, 0.679, 0.653, 0.685, 0.675, 0.682, 0.682, 0.702, 0.697, 0.708, 0.697, 0.722, 0.717, 0.711, 0.705, 0.717, 0.742, 0.719, 0.728, 0.715, 0.733, 0.72, 0.721, 0.715, 0.746, 0.741, 0.734]),
'PIP2': np.array([0.991, 0.987260044, 0.985, 0.982739956, 0.979, 0.973655131, 0.972, 0.977889519, 0.984, 0.983786795, 0.985, 0.994463303, 1.001, 0.994359992, 0.986, 0.988471727, 0.997, 1.001878098, 0.991, 0.95064088, 0.863, 0.721808383, 0.571, 0.455625586, 0.375, 0.317064271, 0.27, 0.225742331, 0.191, 0.172091407, 0.159, 0.142142043, 0.13, 0.131215422, 0.137, 0.137496269, 0.136, 0.137674501, 0.142, 0.147055726, 0.151, 0.152852594, 0.155, 0.159908897, 0.167, 0.174886819, 0.182, 0.187293828, 0.192, 0.19681287, 0.198, 0.193579691, 0.193, 0.204868367, 0.223, 0.238571841, 0.247, 0.24784427, 0.253, 0.27130108, 0.287, 0.285451412, 0.281, 0.289393273, 0.303, 0.311725495, 0.319, 0.329204745, 0.337, 0.338330524, 0.344, 0.363473159, 0.386, 0.395776841, 0.377]),
'IP3': np.array([0, 0, 0, 0, 0.020173079, 0.045400514, 0.012211553, 0.000322887, 0.010325055, 0.030511855, 0.044581874, 0.054724416, 0.068358943, 0.064446759, 0.055239999, 0.033072461, 0.013131853, 0.000866517, 0, 0, 0, 0.014970792, 0.2898243, 0.752767783, 0.930383958, 0.961510528, 0.999909621, 0.96142814, 0.906313748, 0.927474172, 0.966169559, 0.974910744, 0.977399169, 0.940000652, 0.951186253, 0.921090191, 0.833836074, 0.787703727, 0.812838593, 0.812291835, 0.80938294, 0.734586375, 0.676045647, 0.642574059, 0.616124078, 0.593201146, 0.572566747, 0.550100919, 0.53245906, 0.512246625, 0.489295962, 0.471909516, 0.458368649, 0.487777547, 0.45786329, 0.385951806, 0.375624462, 0.38903272, 0.370521362, 0.334801694, 0.386746725, 0.428079727, 0.28953423, 0.278087256, 0.291503843, 0.329827114, 0.229275314, 0.251684177, 0.354957011, 0.31487703, 0.198784886, 0.175409618, 0.212764316, 0.242300052, 0.241267038]),
'stimulus_duration': 20,
'PI_PIP_ratio': 35.00
},
'Neuroblastoma': {
'time': np.array([0, 80, 88, 140, 170, 200, 256, 440, 737, 1038, 1945]),
'PIP': np.array([0.9352, 0.9352, 0.411, 0.392, 0.415, 0.545, 0.6548, 0.7976, 0.975, 1.0, 0.9945]),
'PIP2': np.array([0.9439, 0.9439, 0.2783, 0.2414, 0.671, 0.777, 0.8466, 0.9351, 1.0, 0.986, 0.9985]),
'IP3': np.array([0, 0, 0.7293, 1, 0.7289, 0.5331, 0.3708, 0.1885, 0.1034, 0.091, 0.0341]),
'stimulus_duration': 60,
'PI_PIP_ratio': 22.42
}
}
time_model = np.linspace(0, 300, 100)
for ct, data in experimental_data.items():
for key in ['PIP', 'PIP2', 'IP3']:
experimental_data[ct][key] = np.interp(time_model, data['time'], data[key])
print('Experimental data loaded with known ratios:')
for ct, data in experimental_data.items():
print(f' {ct:15s}: PI/PIP ratio = {data["PI_PIP_ratio"]}')
# %%
model.load_state_dict(torch.load(f'{CHECKPOINT_DIR}/best_model.pth'))
model.eval()
param_names = ['k_PI4K', 'k_PIP5K', 'k_5P', 'k_plc', 'k_deg']
predictions = {}
print('=' * 80)
print('PREDICTIONS vs SCG BASELINE')
print('=' * 80)
for cell_type, data in experimental_data.items():
ratio = data['PI_PIP_ratio']
stim_duration = data['stimulus_duration']
stim_ch = np.full(100, stim_duration / 100.0)
ratio_ch = np.full(100, normalize_ratio(ratio))
X_exp = np.stack([data['PIP'], data['PIP2'], data['IP3'],
stim_ch, ratio_ch], axis=1)
X_exp = torch.FloatTensor(X_exp).unsqueeze(0).to(device)
with torch.no_grad():
y_pred = 10 ** model(X_exp).cpu().numpy()[0]
# Derive k_4P from known ratio and predicted k_PI4K
k_4P_pred = y_pred[0] * ratio
pred_dict = {name: y_pred[i] for i, name in enumerate(param_names)}
pred_dict['PI_PIP_ratio'] = ratio # known, not predicted
pred_dict['k_4P_derived'] = k_4P_pred
predictions[cell_type] = pred_dict
print(f'\n{cell_type} (ratio = {ratio}):')
print('-' * 80)
for name in param_names:
base = baseline[name]
pred = pred_dict[name]
print(f' {name:15s}: Pred = {pred:.6f}, Base = {base:.6f}, Fold = {pred/base:.2f}x')
base_k4P = baseline['k_PI4K'] * baseline['PI_PIP_ratio']
print(f' {"k_4P (derived)":15s}: Pred = {k_4P_pred:.6f}, '
f'Base = {base_k4P:.6f}, Fold = {k_4P_pred/base_k4P:.2f}x')
print('\n' + '=' * 80)
print('Predictions complete!')
# %% [markdown]
# ## Step 1: Network Predictions (above)
#
# ---
# %%
from scipy.optimize import minimize
param_names_5 = ['k_PI4K', 'k_PIP5K', 'k_5P', 'k_plc', 'k_deg']
cell_geometry = {
'SCG': {'PI': 226975, 'SA': 4100, 'Vol': 6644},
'Hippocampal': {'PI': 140000, 'SA': 4100, 'Vol': 6644},
'tsA': {'PI': 140000, 'SA': 1500, 'Vol': 2500},
'Neuroblastoma': {'PI': 17222, 'SA': 700, 'Vol': 700},
}
cell_PIP0 = {
'SCG': 4540,
'Hippocampal': 26000,
'tsA': 4000,
'Neuroblastoma': 768,
}
cell_PIP2_0 = {
'SCG': 3232,
'Hippocampal': 2400,
'tsA': 5000,
'Neuroblastoma': 573,
}
t_model = np.linspace(0, 300, 100)
STIM_IDX = int(80 / 300 * 100)
PARAM_FLOORS = np.array([0.0003, 1e-4, 0.008, 0.05, 1e-4])
PARAM_CEILINGS = np.array([0.002, 0.08, 0.08, 0.6, 0.15])
def simulate(params_vec, ratio, PI, cf, stim_duration, PIP0, PIP2_0):
k_PI4K, k_PIP5K, k_5P, k_plc, k_deg = params_vec
k_4P = ratio * k_PI4K
pdict = {'k_PI4K': k_PI4K, 'k_4P': k_4P, 'k_PIP5K': k_PIP5K,
'k_5P': k_5P, 'k_plc': k_plc, 'k_deg': k_deg}
y0 = [PIP0, PIP2_0, 0.0, 6.0, 0.0]
try:
sol = odeint(pi_cycle_odes, y0, t_model,
args=(pdict, PI, cf, 80, 80 + stim_duration, 0.7),
rtol=1e-8, atol=1e-10)
if not np.isfinite(sol).all():
return None
PIP = sol[:, 0] / sol[0, 0]
PIP2 = sol[:, 1] / sol[0, 1]
mx = sol[:, 4].max()
IP3 = sol[:, 4] / mx if mx > 0 else sol[:, 4]
return PIP, PIP2, IP3
except Exception:
return None
def objective(params_vec, ratio, PI, cf, stim_duration, PIP0, PIP2_0,
exp_PIP, exp_PIP2, exp_IP3,
w_PIP=1.0, w_PIP2=1.0, w_IP3=0.5):
p = np.array(params_vec)
if np.any(p < PARAM_FLOORS) or np.any(p > PARAM_CEILINGS):
return 1e6
result = simulate(p, ratio, PI, cf, stim_duration, PIP0, PIP2_0)
if result is None:
return 1e6
sim_PIP, sim_PIP2, sim_IP3 = result
loss = (w_PIP * np.mean((sim_PIP[STIM_IDX:] - exp_PIP[STIM_IDX:]) ** 2) +
w_PIP2 * np.mean((sim_PIP2[STIM_IDX:] - exp_PIP2[STIM_IDX:]) ** 2) +
w_IP3 * np.mean((sim_IP3[STIM_IDX:] - exp_IP3[STIM_IDX:]) ** 2))
return loss
refined_predictions = {}
print('=' * 80)
print('STEP 1: REFINED PREDICTIONS (Nelder-Mead, biological ICs)')
print('=' * 80)
for cell_type, data in experimental_data.items():
print(f'\nOptimizing {cell_type}...')
pred = predictions[cell_type]
ratio = pred['PI_PIP_ratio']
geo = cell_geometry[cell_type]
PI = geo['PI']
cf = (geo['SA'] / geo['Vol']) * (1 / 602)
sd = data['stimulus_duration']
PIP0 = cell_PIP0[cell_type]
PIP2_0 = cell_PIP2_0[cell_type]
exp_PIP = data['PIP']
exp_PIP2 = data['PIP2']
exp_IP3 = data['IP3']
x0 = np.array([pred[name] for name in param_names_5])
loss_before = objective(x0, ratio, PI, cf, sd, PIP0, PIP2_0,
exp_PIP, exp_PIP2, exp_IP3)
result = minimize(
objective, x0,
args=(ratio, PI, cf, sd, PIP0, PIP2_0, exp_PIP, exp_PIP2, exp_IP3),
method='Nelder-Mead',
options={'maxiter': 5000, 'xatol': 1e-8, 'fatol': 1e-8, 'adaptive': True}
)
x_ref = result.x
k_4P_refined = x_ref[0] * ratio
refined_dict = {name: x_ref[i] for i, name in enumerate(param_names_5)}
refined_dict['PI_PIP_ratio'] = ratio
refined_dict['k_4P_derived'] = k_4P_refined
refined_dict['PIP0'] = PIP0
refined_dict['PIP2_0'] = PIP2_0
refined_dict['PI'] = PI
refined_dict['cf'] = cf
refined_predictions[cell_type] = refined_dict
print(f'{cell_type} (ratio={ratio}, PI={PI}, PIP0={PIP0}, PIP2_0={PIP2_0})')
print('-' * 80)
print(f' {"Parameter":15s} {"Network":>12s} {"Refined":>12s} {"Change":>10s}')
print('-' * 80)
for i, name in enumerate(param_names_5):
print(f' {name:15s} {x0[i]:12.6f} {x_ref[i]:12.6f} {x_ref[i]/x0[i]:10.2f}x')
print(f' {"k_4P (derived)":15s} {pred["k_4P_derived"]:12.6f} {k_4P_refined:12.6f} {k_4P_refined/pred["k_4P_derived"]:10.2f}x')
print(f' Loss before: {loss_before:.6f} -> Loss after: {result.fun:.6f} '
f'(improvement: {(1 - result.fun/loss_before)*100:.1f}%)')
print('\n' + '=' * 80)
print('Refinement complete!')
# %% [markdown]
# ## Step 2: Local Refinement
#
# Takes the network predictions as starting point and runs L-BFGS-B to minimize trajectory MSE directly against experimental data. Bounds are +-60% around each network prediction so the optimizer stays in a physically reasonable region.
#
# The refined plot shows network prediction (gray dashed) vs refined (solid color) vs experimental (dots).
# %%
ss_corrected = {}
print('=' * 80)
print('STEP 2: SS CORRECTION (analytical, k_PIP5K only)')
print('=' * 80)
for cell_type in experimental_data:
ref = refined_predictions[cell_type]
PIP0 = ref['PIP0']
PIP2_0 = ref['PIP2_0']
ratio = ref['PI_PIP_ratio']
PI = ref['PI']
cf = ref['cf']
k_PI4K = ref['k_PI4K']
k_5P = ref['k_5P']
k_plc = ref['k_plc']
k_deg = ref['k_deg']
k_PIP5K_ss = k_5P * (PIP2_0 / PIP0)
k_4P_ss = ratio * k_PI4K
dPIP2_dt = k_PIP5K_ss * PIP0 - k_5P * PIP2_0
dPIP_dt = k_PI4K*PI - k_4P_ss*PIP0 - k_PIP5K_ss*PIP0 + k_5P*PIP2_0
ss_corrected[cell_type] = {
'k_PI4K': k_PI4K,
'k_4P': k_4P_ss,
'k_PIP5K': k_PIP5K_ss,
'k_5P': k_5P,
'k_plc': k_plc,
'k_deg': k_deg,
'PI_PIP_ratio': ratio,
'PIP0': PIP0,
'PIP2_0': PIP2_0,
'PI': PI,
'cf': cf,
}
print(f'\n{cell_type}:')
print(f' {"Parameter":15s} {"Refined":>12s} {"SS-Corrected":>14s}')
print('-' * 50)
print(f' {"k_PI4K":15s} {ref["k_PI4K"]:12.6f} (unchanged)')
print(f' {"k_PIP5K":15s} {ref["k_PIP5K"]:12.6f} {k_PIP5K_ss:14.6f}')
print(f' {"k_5P":15s} {ref["k_5P"]:12.6f} (unchanged)')
print(f' {"k_plc":15s} {ref["k_plc"]:12.6f} (unchanged)')
print(f' {"k_deg":15s} {ref["k_deg"]:12.6f} (unchanged)')
print(f' {"k_4P":15s} {ref["k_4P_derived"]:12.6f} {k_4P_ss:14.6f}')
print(f' dPIP2/dt at t=0 = {dPIP2_dt:.8f} (should be 0)')
print(f' dPIP/dt at t=0 = {dPIP_dt:.8f} (should be ~0)')
print('\n' + '=' * 80)
print('SS correction complete!')
# %%
def plot_final(cell_type):
ref = ss_corrected[cell_type]
pred = predictions[cell_type]
PI = ref['PI']
cf = ref['cf']
PIP0 = ref['PIP0']
PIP2_0 = ref['PIP2_0']
ratio = ref['PI_PIP_ratio']
sd = experimental_data[cell_type]['stimulus_duration']
params_final = {
'k_PI4K': ref['k_PI4K'], 'k_4P': ref['k_4P'],
'k_PIP5K': ref['k_PIP5K'], 'k_5P': ref['k_5P'],
'k_plc': ref['k_plc'], 'k_deg': ref['k_deg'],
}
params_net = {
'k_PI4K': pred['k_PI4K'], 'k_4P': pred['k_4P_derived'],
'k_PIP5K': pred['k_PIP5K'], 'k_5P': pred['k_5P'],
'k_plc': pred['k_plc'], 'k_deg': pred['k_deg'],
}
y0_final = [PIP0, PIP2_0, 0.0, 6.0, 0.0]
y0_net = [PIP0, (pred['k_PIP5K'] / pred['k_5P']) * PIP0, 0.0, 6.0, 0.0]
t = np.linspace(0, 300, 100)
sol_final = odeint(pi_cycle_odes, y0_final, t,
args=(params_final, PI, cf, 80, 80+sd, 0.7),
rtol=1e-8, atol=1e-10)
sol_net = odeint(pi_cycle_odes, y0_net, t,
args=(params_net, PI, cf, 80, 80+sd, 0.7),
rtol=1e-8, atol=1e-10)
def norm(sol, col):
v = sol[:, col]
return v / v[0] if v[0] != 0 else v
def norm_ip3(sol):
v = sol[:, 4]
mx = v.max()
return v / mx if mx > 0 else v
fig, axes = plt.subplots(1, 3, figsize=(18, 5))
for ax, (name, col, color), ekey in zip(
axes,
[('PIP', 0, 'black'), ('PIP2', 1, 'red'), ('IP3', None, 'magenta')],
['PIP', 'PIP2', 'IP3']
):
exp = experimental_data[cell_type][ekey]
n_trace = norm(sol_net, col) if col is not None else norm_ip3(sol_net)
f_trace = norm(sol_final, col) if col is not None else norm_ip3(sol_final)
ax.scatter(t, exp, color=color, alpha=0.5, s=30, label='Experimental', zorder=3)
ax.plot(t, n_trace, color='gray', linewidth=2, linestyle='--', label='Network', zorder=1)
ax.plot(t, f_trace, color=color, linewidth=3, label='Final', zorder=2)
ax.axvline(80, color='blue', linewidth=1, linestyle=':', alpha=0.6, label='Stim on')
ax.axvline(80+sd, color='blue', linewidth=1, linestyle='-.', alpha=0.6, label='Stim off')
ax.set_xlabel('Time (s)', fontsize=13, fontweight='bold')
ax.set_ylabel(f'{name} (normalized)', fontsize=13, fontweight='bold')
ax.set_title(f'{cell_type} - {name} (ratio={ratio})', fontsize=14, fontweight='bold')
ax.legend(fontsize=10)
ax.grid(alpha=0.3)
ax.set_xlim([0, 300])
plt.tight_layout()
plt.savefig(f'{OUTPUT_DIR}/{cell_type}_final.png', dpi=150, bbox_inches='tight')
plt.show()
print(f'{cell_type} final plot saved.')
for ct in experimental_data:
plot_final(ct)
# %% [markdown]
# ## Done
#
# **v4 architecture change:**
# - Input channels: PIP, PIP2, IP3, stim_duration, **ratio** (5 total, was 4)
# - Output parameters: k_PI4K, k_PIP5K, k_5P, k_plc, k_deg (5 total, was 6)
# - PI_PIP_ratio is now a **known input** from biology, not a predicted output
# - k_4P derived as `known_ratio * predicted_k_PI4K` at prediction time
# - Plots use cell-specific PI pools and geometry from literature
#
# **Known ratios used:**
# - SCG: 50.00
# - tsA: 35.00
# - Neuroblastoma: 22.42
# - Hippocampal: 5.38
#
# **To retrain from scratch:** uncomment Cell 1, run it, comment it back, Run All.

PI_Cycle_v6.ipynb at commit 3094923, under GPL-3.0 · at the source

Overview

Authors: Gonzalo Hernandez-Hernandez^1,2, Mindy Tieu², Pei-Chi Yang^1,2, Oscar Vivas³, Timothy J Lewis⁴, L Fernando Santana², Colleen E Clancy^1,2,5

ORCID iDs: Colleen E Clancy

Center for Precision Medicine and Data Science, University of California, Davis, CA, USA
University of California Davis, Department of Physiology and Membrane Biology, Davis, CA, USA
Department of Physiology and Biophysics, University of Washington, Seattle, WA, USA
Department of Mathematics, University of California, Davis, CA, USA
Department of Pharmacology, University of California, Davis, CA, USA

Institutions: University of California, Davis (United States); University of Washington (United States)

Journal: Journal of precision medicine (Amsterdam, Netherlands), volume 7, article 100050

Dates: published online 25 July 2026; in print September 2026

Type: Research article · Language: English

License: CC BY-NC-ND

Identifiers: DOI 10.1016/j.premed.2026.100050 · PMID 42741463 · PMCID PMC13574257 · OpenAlex W7168787831

Open access: hybrid, a free copy (OpenAlex)

Status: code verified

Methods: Machine learning

Keywords: Phosphoinositide signaling, New approach methodologies (NAMs), Precision medicine, Biological sciences, Biophysics and computational biology

Topic: Protein Kinase Regulation and GTPase Signaling (Molecular Biology, Biochemistry, Genetics and Molecular Biology), according to OpenAlex

Funding: NHLBI NIH HHS (R01 HL174001, R01 HL168874, R01 HL128537); NIGMS NIH HHS (R35 GM142690)

Citations: not cited yet (Europe PMC); 31 references in the paper

Abstract

The abstract is not reproduced here: the paper's license (CC BY-NC-ND) does not allow it. Read it in the paper, at the publisher or on Europe PMC.

Repository

Its files are read in the Code ↔ Paper reader above, with 11 matches between paragraphs and lines of code.

ClancyLabUCD/from-lipid-dynamics-to-precision-predictions

License: GPL-3.0

State: the link answers, verified on 27 September 2026

Evidence: files inventoried

Commit: 30949231ef04ff484224489768dfd0ef6ca059fa, 29 June 2026

Languages: MATLAB (20), Jupyter (1)

Size: 23 files, 21 scripts

Software Heritage: not archived

Found in: the text, “Simulation protocols”

Holds: README, license file, 1 notebook

Not found: CITATION.cff, environment file, tests, continuous integration, documentation

Tools: Parallel Computing Toolbox (2 files), Matplotlib (1 file), NumPy (1 file), PyTorch (1 file), scikit-learn (1 file), SciPy (1 file)

Availability: 1 check, the latest on 27 September 2026: the link answers

27 September 2026: the link answers

23 files

Codes_For_Figures/Experimental_Data/PIP2_Validation_SCG.m, MATLAB, 28 lines
Codes_For_Figures/Experimental_Data/z_data_Hippocampal.m, MATLAB, 49 lines
Codes_For_Figures/Experimental_Data/z_data_Neuroblastoma.m, MATLAB, 50 lines
Codes_For_Figures/Experimental_Data/z_data_SCG.m, MATLAB, 49 lines
Codes_For_Figures/Experimental_Data/z_data_tsA_2016.m, MATLAB, 47 lines
Codes_For_Figures/Experimental_Data/z_data_tsA_2022.m, MATLAB, 46 lines
Codes_For_Figures/Figure_2/Figure_2_A_Optimization/all_fitcurve.m, MATLAB, 63 lines
Codes_For_Figures/Figure_2/Figure_2_A_Optimization/fminsearchbnd.m, MATLAB, 217 lines
Codes_For_Figures/Figure_2/Figure_2_A_Optimization/gate_functions.m, MATLAB, 138 lines
Codes_For_Figures/Figure_2/Figure_2_A_Optimization/main_optimization.m, MATLAB, 147 lines, 1 match
Codes_For_Figures/Figure_2/Figure_2_B.m, MATLAB, 225 lines, 1 match
Codes_For_Figures/Figure_2/Isolated_SCG_Neurons.m, MATLAB, 200 lines, 1 match
Codes_For_Figures/Figure_3/Figure_3_ABC/Local_Sensitivity_Analysis.m, MATLAB, 545 lines
Codes_For_Figures/Figure_4/Global_Sensitivity_Analysis.m, MATLAB, 671 lines, 2 matches
Codes_For_Figures/Figure_6/Colab/PI_Cycle_v6.ipynb, Jupyter, 767 lines, 4 matches
Codes_For_Figures/Figure_6/Figure6_CDE_NN.m, MATLAB, 219 lines
Codes_For_Figures/Figure_6/Figure_6_AB.m, MATLAB, 132 lines, 2 matches
Codes_For_Figures/Figure_7/SCG_Neurons_PI4K_Combined_3D_View.m, MATLAB, 243 lines
Codes_For_Figures/Figure_7/SCG_Neurons_PIP5K_Combined_3D_View.m, MATLAB, 248 lines
Codes_For_Figures/Figure_8/Hippocampal_Neurons_PI4K_Combined_View.m, MATLAB, 187 lines
Codes_For_Figures/Figure_8/Hippocampal_Neurons_PIP5K_Combined_View.m, MATLAB, 192 lines
LICENSE, License, 674 lines
README.md, Text, 82 lines

At the source: github.com/ClancyLabUCD/from-lipid-dynamics-to-precision-predictions

The paper's code and data availability statement is in the Data section.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
21 scripts, each with its path and the digest of its content;
11 matches between paragraphs of the paper and lines of the code (method lexical-v1);
neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

No dataset and no data link were found in the paper.

Code and data availability statement

The paper has a code and data availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

it says that the data are available on request

Read it in the paper: doi.org/10.1016/j.premed.2026.100050.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 2, 28 September 2026

Publisher: n/a → Elsevier BV

Version 1, 27 September 2026: the first record

Recorded: type, language, journal, volume, pages, dates, 7 authors, 5 keywords, 2 funders, 31 references.

Cite

This paper

Hernandez-Hernandez, G., Tieu, M., Yang, P.-C., Vivas, O., Lewis, T. J., Santana, L. F., & Clancy, C. E. (2026). From lipid dynamics to precision predictions: A new approach methodology for precision modeling of phosphoinositide signaling. Journal of precision medicine (Amsterdam, Netherlands), 7, 100050. https://doi.org/10.1016/j.premed.2026.100050

BibTeX

@article{hernandezhernandez2026lipid, author = {Hernandez-Hernandez, Gonzalo and Tieu, Mindy and Yang, Pei-Chi and Vivas, Oscar and Lewis, Timothy J and Santana, L Fernando and Clancy, Colleen E}, title = {{From lipid dynamics to precision predictions: A new approach methodology for precision modeling of phosphoinositide signaling}}, journal = {Journal of precision medicine (Amsterdam, Netherlands)}, year = {2026}, month = jul, volume = {7}, pages = {100050}, publisher = {Elsevier BV}, issn = {3050-6328}, doi = {10.1016/j.premed.2026.100050}, url = {https://doi.org/10.1016/j.premed.2026.100050}, pmid = {42741463}, pmcid = {PMC13574257} }

RIS

TY - JOUR AU - Hernandez-Hernandez, Gonzalo AU - Tieu, Mindy AU - Yang, Pei-Chi AU - Vivas, Oscar AU - Lewis, Timothy J AU - Santana, L Fernando AU - Clancy, Colleen E TI - From lipid dynamics to precision predictions: A new approach methodology for precision modeling of phosphoinositide signaling T2 - Journal of precision medicine (Amsterdam, Netherlands) J2 - J Precis Med (Amst) PY - 2026 DA - 2026/07/25 VL - 7 SP - 100050 SN - 3050-6328 PB - Elsevier BV DO - 10.1016/j.premed.2026.100050 UR - https://doi.org/10.1016/j.premed.2026.100050 LA - en ER -

CSL-JSON

{ "id": "10.1016/j.premed.2026.100050", "type": "article-journal", "title": "From lipid dynamics to precision predictions: A new approach methodology for precision modeling of phosphoinositide signaling", "container-title": "Journal of precision medicine (Amsterdam, Netherlands)", "author": [ { "family": "Hernandez-Hernandez", "given": "Gonzalo" }, { "family": "Tieu", "given": "Mindy" }, { "family": "Yang", "given": "Pei-Chi" }, { "family": "Vivas", "given": "Oscar" }, { "family": "Lewis", "given": "Timothy J" }, { "family": "Santana", "given": "L Fernando" }, { "family": "Clancy", "given": "Colleen E" } ], "container-title-short": "J Precis Med (Amst)", "volume": "7", "page": "100050", "DOI": "10.1016/j.premed.2026.100050", "PMID": "42741463", "PMCID": "PMC13574257", "ISSN": "3050-6328", "publisher": "Elsevier BV", "URL": "https://doi.org/10.1016/j.premed.2026.100050", "language": "en", "issued": { "date-parts": [ [ 2026, 7, 25 ] ] } }

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.