Release Notes — Polyauxic Modeling Platform v1.0.0 (Streamlit) Release date: 2025-12-22 Scope: First public, end-to-end release of a trilingual (EN/PT/FR) Streamlit application for mono- and polyauxic sigmoidal modeling with robust fitting, outlier handling, and principled model selection. What's new (v1.0.0) Core capabilities End-to-end kinetic modeling workflow: upload → parse replicates → fit 1..N phases → select best phase count → visualize and export plots. Two-model engine implementing the platform's reparameterized sigmoids: Boltzmann (Eq. 31) Gompertz (Eq. 32) with polyauxic growth as a weighted sum of phases, where weights are constrained via Softmax (stable, identifiable phase weights). Robust fitting (global → local) Stage 1: Differential Evolution (DE) for global exploration (helps avoid local minima in multi-phase landscapes). Stage 2: L-BFGS-B for constrained local refinement (bounds enforced for stability and biological plausibility). Outlier handling (user-selectable) No removal: uses all points. ROUT-like (Simple MAD): robust z-score filtering (|Z| > 2.5) based on MAD scale. ROUT (Robust + FDR): robust pre-fit + p-values + Benjamini–Hochberg FDR control with user-defined Q (%). Model selection with parsimony Fits 1 to Max Phases and reports AIC / AICc / BIC, plus R² / adjusted R² / SSE. Automatically chooses which information criterion to prioritize (AIC vs AICc vs BIC) based on sample size and parameter-to-data ratio logic. Selects the first local minimum of the chosen criterion (prevents "phase inflation" even if later phase counts look numerically better). Uncertainty quantification Standard errors estimated via a numerical Hessian and residual variance (with pseudo-inverse fallback). Phase-weight uncertainty propagated from Softmax parameters using a Jacobian-based approach (delta-method style). User experience and UI Trilingual UI (🇬🇧 / 🇧🇷 / 🇫🇷) for titles, instructions, sidebar controls, plots, and tables. Replicate-aware ingestion: expects paired columns (t1,y1,t2,y2, …) and supports up to 5 biological replicates. Variable presets: generic y(t), product P(t), substrate S(t), biomass X(t) (labels + rate symbol mapping). Constraints panel: Force yᵢ = 0 Force y𝒻 = 0 (disabled when yᵢ is forced for safety) Outputs Per-fit plots (global + phase decomposition) with SVG download. Criteria summary chart (metrics vs number of phases) with SVG download. Summary tables for: global parameters (yᵢ, y𝒻), phase parameters (p, r_max, λ) + SEs, fit metrics (R², adjusted R², AIC/AICc/BIC). Known limitations (v1.0.0) Computation time scales quickly with phase count because DE is expensive (expected for global optimization). ROUT implementation is "ROUT-style" in practice (robust scale + FDR), not a byte-for-byte reproduction of any single proprietary implementation; treat it as statistically-motivated screening, not ground truth. Information-criterion rule is intentionally conservative (first local minimum). If your use-case truly needs maximum-phase fitting, you'll want a "global minimum" toggle (not included in v1.0.0). Scientific basis / suggested citation Mockaitis, G. (2025). Mono and Polyauxic Growth Kinetic Models. arXiv:2507.05960. Differential Evolution: Storn, R., & Price, K. (1997). Differential Evolution – A Simple and Efficient Heuristic for global Optimization over Continuous Spaces. L-BFGS-B: Byrd, R. H., Lu, P., Nocedal, J., & Zhu, C. (1995). A Limited Memory Algorithm for Bound Constrained Optimization. ROUT: Motulsky, H. J., & Brown, R. E. (2006). Detecting outliers when fitting data with nonlinear regression – a new method based on robust regression and false discovery rate. FDR: Benjamini, Y., & Hochberg, Y. (1995). Controlling the false discovery rate: a practical and powerful approach to multiple testing. AIC: Akaike, H. (1974). A new look at the statistical model identification. AICc: Hurvich, C. M., & Tsai, C.-L. (1989). Regression and time series model selection in small samples. BIC: Schwarz, G. (1978). Estimating the dimension of a model.