AroMIP — Aromatic Membrane Insertion Predictor

AroMIP predicts whether aromatic-centered motifs in intrinsically disordered proteins (IDPs) insert into the acyl-chain region of lipid membranes — directly in your browser, no installation required.

---

Overview

Intrinsically disordered proteins (IDPs) frequently interact with cellular membranes through short hydrophobic motifs centered on aromatic residues (Phe, Trp, Tyr). These membrane-inserting motifs play key roles in signaling, aggregation, and pathological processes. AroMIP provides a fast, interpretable score for each such motif in any query sequence.

---

How It Works

1. Fragment Scanning

The tool scans the input sequence and extracts every 9-residue window whose center residue (position 5) is an aromatic amino acid — F, W, or Y.

2. Multiplicative Scoring Model

Each flanking residue (positions 1–4 and 6–9) contributes a multiplicative factor derived from training on intrinsically disordered regions of the human proteome. The raw insertion propensity Q is the product of all eight flanking factors for the central aromatic type:

$$Q = \prod_{\substack{i=1 \ i \neq 5}}^{9} \alpha_i^{(\text{AA}_i)}$$

where $\alpha_i^{(\text{AA}_i)}$ is the trained weight for the amino acid at position i, specific to the central aromatic (F, W, or Y).

3. Generalized Sigmoid

Q is converted to a membrane-insertion propensity score P ∈ [0, 1) via a generalized sigmoid:

$$P = \begin{cases} \dfrac{Q}{1 + Q} & \text{if } Q > 0 \ 0 & \text{otherwise} \end{cases}$$

4. Classification

Score P	Prediction
> 0.5	Inserter
≤ 0.5	Non-inserter

---

Using the Web Tool

1. Open the live demo.
2. Paste or type your protein sequence (single-letter amino acid codes) into the text box.
3. Click Predict! — results appear instantly in a table below.
4. Use Input Example to try the demo sequence RRNKFGINRTTGNWRGMLQRDLYSGLN.

Each row in the results table shows:

Column	Description
Motif	9-residue fragment centered on the aromatic residue
Membrane-insertion propensity (P)	Score from 0 to 1
Prediction	Inserter or Non-inserter

---

Example

Input sequence:

RRNKFGINRTTGNWRGMLQRDLYSGLN

Expected output includes motifs centered on F, W, and Y, each scored and classified. The model correctly captures the hydrophobic context favoring membrane insertion.

---

Model Details

The model parameters were trained separately for F, W, and Y as the central aromatic residue. Residues with strong positive weights (> 1) in the flanking positions — such as Ile, Leu, Val, Met, and other aromatics — generally promote membrane insertion, while charged and polar residues (Asp, Glu, Asn, Gln) suppress it.

---

Training Code & Data

The model parameters and datasets used to build AroMIP are available in the codes/ folder:

• codes/5letter/ — training convergence study on a fully enumerated synthetic dataset (5-letter reduced alphabet, ~390 k sequences per aromatic)
• codes/HumanProteomeIDRs/ — train/test scripts on 9-residue windows from human proteome IDRs (full 20-AA alphabet, pre-split 80/20)

See codes/README.md for full details on the model, datasets, and how to run the scripts.

---

Citation

If you use AroMIP in your research, please cite:

Fidha Kunnath Muhammedkutty and Huan-Xiang Zhou (2026).
A membrane insertion code for intrinsically disordered proteins.
bioRxiv.

---

Author

Fidha Nazreen Kunnath Muhammedkutty
PhD Candidate, Department of Chemistry
University of Illinois Chicago