What is this AI model for?

This portal predicts how likely a small molecule is to show activity against the three most-studied protein targets in Alzheimer's drug discovery:

How it works: three classifiers (deep neural network, histogram gradient boosting, random forest) each score your molecule from its structure — a 2048-bit ECFP4 fingerprint plus 10 physicochemical descriptors — and a meta-learner combines them into the Multi-Target Composite Score (MTCS). The ensemble was trained on curated anti-Alzheimer bioactivity data from the PhD study of Evolvulus alsinoides and Cinnamomum (94.20% accuracy, ROC-AUC 0.958 — see the About tab).

How to use it: enter a molecule as a SMILES string (or click a preset below), press Predict, and read the results:

The descriptors panel shows drug-likeness (Lipinski, Veber) and brain-penetration potential (CNS MPO, BBB category). Score many molecules at once in the Batch tab (with CSV export), browse known reference compounds with literature citations in the Library, and inspect the target proteins in 3D in the Viewer.

Predictions are computational estimates that help you prioritize candidates for further study — they are not experimental proof of activity. Always follow up with docking, MD simulation or wet-lab assays.

AChE, BACE1 & GSK-3β Inhibitor Prediction from SMILES

The free AI tool for multi-target directed ligand (MTDL) design: paste a SMILES string, get predicted inhibition probabilities for all three targets, CNS drug-likeness descriptors and a composite MTCS lead score in seconds.

Using this model in Alzheimer's research — a researcher's plan

Where it fits: the predictor is the first, cheapest filter in the modern in-silico drug-discovery funnel. Because single-target Alzheimer's drugs have repeatedly failed in trials, current research favors multi-target ligand design (MTDL) — one molecule acting on AChE, BACE1 and/or GSK-3b together. This model exists precisely to find such multi-target candidates before you spend compute or lab budget on them: library → AI triage (here) → molecular docking → MD simulation → in-vitro assays.

  1. Frame the hypothesis. Decide your strategy: multi-target (use the MTCS score as the headline metric) or focused on one target (read that target's probability, use MTCS as supporting evidence). State it before screening — in a thesis, this becomes your objective.
  2. Assemble the candidate library. Typical sources: phytochemicals of a medicinal plant (e.g. from IMPPAT), repurposed approved drugs, or designed analogs of a known scaffold. 50–500 compounds is a practical starting set.
  3. Sanity-check with controls. Before trusting the screen, run known actives and inactives: Donepezil (positive control for AChE — it is in the presets) and a non-CNS molecule such as aspirin. The model should separate them clearly.
  4. Batch screen and filter. Paste the library in the Batch tab. Tier 1 (Lipinski + Veber + CNS MPO ≥ 3) removes compounds that are not drug-like or cannot reach the brain; MTCS then ranks the survivors. Export the CSV for your records.
  5. Select candidates wisely. Take the top MTCS scorers and a few structurally diverse moderate scorers — scaffold diversity hedges against model bias. Record selection criteria; examiners ask.
  6. Validate computationally. Dock the selected candidates against the crystal structures (the five targets are loadable in the 3D Viewer tab; full docking runs on the BioDockify platform), then submit the best complexes to MD simulation for pose stability.
  7. Plan experimental follow-up. Ellman's assay for AChE inhibition, FRET-based assay for BACE1, kinase assay for GSK-3b — then cell-line studies. Computational predictions prioritize; experiments prove.

For publication, cite the underlying study (see About), report the model version and access date, and always present AI predictions alongside docking and experimental evidence — never as standalone proof.

Questions researchers ask

Can I predict AChE, BACE1 and GSK-3β inhibition from SMILES in one step?
Yes — that is exactly what this tool does. Enter a SMILES string and the ensemble returns a probability of inhibition for each of the three targets plus a composite multi-target score, drug-likeness descriptors and a 2D structure.

Is it free, and do I need to install anything?
Completely free and browser-based — no installation, no account. Batch screening of up to 200 molecules and CSV export are included.

How is this different from SwissTargetPrediction or AlzPlatform?
General target-fishing servers rank many possible protein targets for a molecule. This tool is purpose-built for anti-Alzheimer multi-target ligand design: it scores the three targets that matter for the MTDL strategy with a trained ensemble, and adds CNS-specific developability (CNS MPO, BBB category).

Can I screen phytochemicals or plant extracts for anti-Alzheimer activity?
Yes. Paste the SMILES of your phytochemical library (e.g. from IMPPAT) in the Batch tab, filter with Tier 1 drug-likeness, then rank by the multi-target score before docking and MD studies.

Should I trust the predictions for my thesis?
Use them to prioritize candidates — always pair them with molecular docking, MD simulation and experimental assays (Ellman for AChE, FRET for BACE1, kinase assay for GSK-3β), and cite the study when you publish.

How to use the predictor

1Get a molecule

Type or paste a SMILES string, or click a preset compound below the box.

2Run the prediction

The AI ensemble scores activity against AChE, BACE1 and GSK-3b in seconds.

3Read the results

Composite MTCS score, per-target probabilities, drug-likeness descriptors and 2D structure.

Single-Molecule AI Multi-Target Predictor

Enter a canonical SMILES string to predict multi-target activity against AChE, BACE1, and GSK-3 using our stacked ensemble (94.20% accuracy, ROC-AUC 0.958).

Not sure what to enter? Load the Luteolin example or pick any preset compound below.