New features in FormuSense add flexibility for extracting value from your product formulation data.

The “PCA Release” of FormuSense adds the ability to build Principal Component Analysis (PCA) models, encouraging users to explore and understand variation in one or more blocks of input (X) variables, and to run simulations and optimizations on those PCA models.
 
PCA is a complementary modeling methodology to the existing method included in earlier releases of FormuSense (PLS). While PLS models include both input (X) and outcome (Y) variables, PCA models are inherently restricted to inputs (X).

PCA modeling, simulation and optimization can be useful in a variety of situations including:

    • in early stages of exploring data-driven product formulation tools (as one may find that outcome data is unavailable or difficult to obtain),
    • when a particular set of input variables have been identified for in-depth analysis (such as locating gaps in the product recipe design space), and
    • when one wishes to focus analysis solely on product quality/performance variables (typically outcome variables in a PLS model).

Start Your Trial


Download a copy of FormuSense from the FormuSense homepage, and activate your 30-day trial today.
 

Attend a Demo or Workshop

Register for an upcoming demo (typically 45-min) or an interactive product development workshop (2-hours) with ProSensus by browsing the FormuSense events page.

Release Notes

Below is an expandable summary of new features added and changes implemented in FormuSense as part of the version released in February 2024. This version is referred to as the PCA Release, but also includes other updates that improve mixture properties and modeling analysis.

What’s New

Mixture Properties
    • Configuration:
      • Automatically generate a global mixture property for each ingredient property that appears across all ingredient classes (with the exact same nomenclature).
Modeling
    • Configure, prune, and interpret principal component analysis models (PCA). Note that PCA models inherently include inputs (X) but no outputs (Y).
    • Setup:
      • Specify new models as either PLS (X and Y variables) or PCA (only X variables).
      • Ability to include quality variables as inputs (X) to PCA models. Note that FormuSense treats any quality variables included in a PCA model as a separate process variable block.
    • Analysis:
      • Review generated PCA models with three analysis views. Note that PCA analysis views are slightly different than PLS views, and can be customized to include four intuitive visualizations, chosen from eight different options.
      • New model analysis plot (PLS and PCA): Raw Data Scatter Plots – to plot any two X or Y variables against one another in original units.
      • New model analysis plot (PCA): P Bar Plot – shows the x-loadings, which are the optimal weights (P) used to calculate the weighted average (scores).
      • Generate and copy a “Model Information and Comparison” table that summarizes all configured models in terms of number of principle components, number of variables, number of formulations, and model fit parameters.
Simulation
    • Specify input values (X) for any PCA model to project the results onto the latent space.
      • View results: Review tabular and graphical summaries of all PCA simulation results relative to the historical formulations that were used to build the model.
      • When applicable, model validity metrics, formulation cost, and mixture properties are calculated for PCA simulations.
      • Find the historical formulation closest to each PCA simulation scenario.
      • Export: PCA simulation results can be exported to Excel directly from the simulation results screen.
Optimization
    • Specify PCA score space targets to determine the optimal input values (X) to achieve those targets under given constraints.
    • After targets, bound constraints and tuning parameters have been selected and the optimization problem is solved, various tabular and graphical summaries are provided to evaluate PCA solution scenarios that prioritize or balance quality and raw material costs.
    • PCA solution scenarios can also be compared to the closest-matching existing historical product.

What’s Changed

Modeling
    • Analysis:
      • Where font size for datapoint labels on Loadings and Weights plots were previously static, the font size is now dictated by the number of datapoints (with fewer datapoints resulting in a larger font size).