ZIF Biocomposite Explorer

About ZIF Biocomposites

Zeolitic Imidazolate Frameworks (ZIFs) are a class of metal–organic frameworks built from tetrahedral metal nodes (here zinc, Zn²⁺) coordinated by imidazolate linkers. Their porous, cage-like topology gives ZIFs exceptional chemical stability and a tunable internal surface area, making them attractive hosts for bioactive cargo.

In this study, ZIF biocomposites are formed by mixing zinc nitrate, 2-methylimidazole (Hmim), and Bovine Serum Albumin (BSA) in aqueous solution. BSA serves as both a structural co-precipitant and a model encapsulant; its presence during ZIF nucleation profoundly alters the crystalline phase that forms.

The resulting composites are of interest for protein delivery, biocatalysis, and biosensing because the ZIF shell can protect labile biomolecules from enzymatic degradation while the pore structure allows controlled release.

A central challenge is the simultaneous competition between multiple ZIF polymorphs during synthesis. The outcome depends strongly on the metal-to-ligand ratio, total reagent concentration, the amount of protein present, and the post-synthesis washing protocol.

Synthesis & Characterisation

Samples were prepared by combining zinc nitrate (Metal), 2-methylimidazole (Ligand), and BSA in varying molar ratios at five total concentration levels: 12.5, 25, 50, 75, and 100 mg·mL^⁻¹. Each formulation was prepared in triplicate to quantify measurement reproducibility.

After precipitation, samples were washed with either ethanol or water to remove unreacted reagents. The two washing protocols yield different surface chemistries and can shift the phase equilibrium of the final solid.

Each sample was characterised by:

Powder XRD — diffraction patterns were matched against reference ZIF structures to assign the dominant crystalline phase and quantify crystallinity (the crystalline fraction of the total scattering volume).

ATR-IR spectroscopy — amide and imidazolate band intensities provide an independent estimate of the protein-to-framework ratio retained in the composite, reported here as the ATR-IR bands ratio.

Phase Identification Guide

XRD patterns were matched against known ZIF structure databases. Eight distinct outcomes are tracked in this dataset, from fully amorphous precipitates to well-defined crystalline polymorphs with different cage topologies:

Amorphous

Sodalite (SOD)

Diamondoid (DIA)

ZIF-C

U13

U12

ZIF-EC-1

ZIF-L

Sodalite is the canonical ZIF-8 topology and typically forms under high-ligand conditions. Diamondoid is a denser polymorph favoured at intermediate ratios. ZIF-L is a leaf-like two-dimensional phase. U12 and U13 are newly identified phases specific to this biocomposite system. Mixed-phase samples display contributions from several of these simultaneously, which the inspector reports as phase probabilities.

Reading the Ternary & 3D Views

A ternary diagram maps three-component mixtures onto a triangle. Each vertex represents 100 % of one component; any interior point encodes all three fractions simultaneously. Because Metal + Ligand + BSA = 100 %, the position of a point fully determines its composition.

Moving towards the Metal vertex (bottom-left) increases the zinc content. Moving towards Ligand (bottom-right) increases the imidazolate ratio. Moving towards BSA (top) increases the protein concentration.

The 3D stacked view layers five such triangles vertically, one per concentration level. The depth axis therefore represents total reagent concentration, revealing how the same composition behaves as the solution is diluted or concentrated.

Machine Learning Predictor

The Prediction data layer uses, when scikit-learn is available, a soft-voting ensemble of Random Forest (RF) and Extra Trees (ET) classifiers trained on the full current Exp-A dataset (~360 samples, 5 observed phase classes). Both models use 300 estimators, balanced class weights, and max_features = sqrt(p). The figures below reflect a local benchmark on the current Exp-A dataset run on 26 May 2026 using 7-fold stratified cross-validation, set by the smallest class count. If scikit-learn is unavailable, the app falls back to a nearest-neighbour phase-classification prototype.

Random Forest

84.7%

± 5.8 pp (7-fold CV)

n_estimators = 300 · class_weight = balanced

Extra Trees

84.1%

± 7.2 pp (7-fold CV)

n_estimators = 300 · class_weight = balanced

RF + ET Ensemble Deployed

84.4%

± 7.9 pp (7-fold CV)

Soft-vote · averaged class probabilities · latest local benchmark on the current Exp-A dataset

6-feature engineered input space

Metal fraction (%), Ligand fraction (%), Total concentration (mg mL⁻¹), log(concentration), Metal/Ligand molar ratio, Washing protocol (ethanol = 1, water = 0). BSA% is excluded because Metal + Ligand + BSA = 100 % makes it perfectly collinear with the other two fractions.

Regression targets & confidence

Encapsulation efficiency, crystallinity, loading capacity, and ATR-IR bands ratio are interpolated with k = 9 nearest-neighbour regression (LOO-CV optimal). Each prediction carries a distance confidence band: Near (< p33), Moderate (p33–p75), or Far (> p75) from the nearest training point. Predictions are restricted to the experimentally supported component and concentration ranges.

Explorer Features

3D Stacked View

All five concentration layers rendered simultaneously as a three-dimensional stack, revealing phase evolution across the concentration axis.

2D Ternary View

Isolates one concentration layer for precise composition–phase analysis without inter-layer visual overlap.

Flexible Coloring

Color markers by dominant phase, crystallinity, encapsulation efficiency, ATR-IR bands ratio, loading capacity, or individual phase probability.

Advanced Filtering

Filter by minimum crystallinity, EE threshold, ATR-IR ratio, phase probability minimums, and composition slices through the ternary space.

Sample Inspector

Click any measured point to open its full record: composition, crystallinity, phase fractions, EE, ATR-IR spectrum, and repeated XRD experiments with CSV download.

Search & Predict

Place a custom composition marker anywhere in the diagram and query the ML ensemble for phase probability estimates at that formulation.