Computational Drug Discovery Targeting a Rare Cystic Fibrosis-linked CFTR Mutant

CWSF · 2026 Disease & Illness Silver Medal

Kavin Rithwik · Senior Division · Grades 11–13

Thumbnail supplied by the source for Computational Drug Discovery Targeting a Rare Cystic Fibrosis-linked CFTR Mutant

Overview

Cystic fibrosis is a disease that causes difficulty breathing and digestive problems because a certain protein in the body is mutated and therefore does not work properly. Using computational software, I searched for potential drugs to treat a rare form of cystic fibrosis that currently has no approved treatments. I used an advanced AI model called AlphaFold3 to predict the 3D shape of the faulty protein and tested its interactions with 300 chemical compounds from a database. The docking software, Autodock Vina, assessed binding strength and ranked several candidates that interact with the mutated protein. Interestingly, one potential drug attached to a part of the protein that current drugs do not target, while others attached to the same domains as existing drugs. If confirmed with lab tests, these molecules could lead to the first ever treatment for people with this rare mutation.

Video

If the video does not play, watch it on the original project page.

Video

Video Transcript

Cystic fibrosis affects over 112,000 people worldwide and is caused by mutations in a channel protein called CFTR. Patients experience difficulty breathing, digestive complications, and frequent infections. Recently, life-changing drugs like Trikafta have been developed for patients with the common ΔF508 mutation. But for those with rare mutations like R560S, there are no approved treatments. Advances in computational methods have promise in significantly accelerating the drug discovery process. I built a 3D model of the faulty R560S mutant using an artificial intelligence protein structure prediction tool called AlphaFold. I then virtually screened drug-like molecules to find six promising candidates. These candidate drugs require laboratory testing, but instead of testing billions of compounds, we can focus on the handful of promising ones. My project shows that computer-based screening can help find drugs for currently untreatable mutations, potentially helping many cystic fibrosis patients in Canada and around the world. Thank you.

Why?

Cystic fibrosis (CF), affecting over 112,000 individuals worldwide [1], is an autosomal recessive disorder caused by mutations in the Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) anion channel protein [2] (Figure 1) [3]. CF causes thick, sticky mucus that clogs the lungs, pancreas, and other organs [2]. This mucus buildup results in chronic infections, difficulty breathing, and digestive complications, with severity varying by mutation type [2]. One in seven CF patients are ineligible for existing medications, and in 2022, over half of Canadian CF patients died before age 40 [4].

CFTR mutations prevent chloride and bicarbonate ions from leaving cells. At the same time, ENaC (a sodium channel normally regulated by CFTR) pumps sodium ions into cells. Increased ion concentration within a cell causes intracellular water accumulation by osmosis, therefore dehydrating extracellular mucus (Figure 2) [5].

Current CFTR drugs contain potentiators (hold the gate open) and correctors (help the protein fold correctly so it can reach the cell surface) [6]. There are over 2,500 known CFTR mutations, with ΔF508 accounting for approximately 70% of cases [7]. Trikafta (Figure 3) is a highly effective therapy for patients with ΔF508, treating about 90% of patients [8]. Unfortunately, those with rare mutations like R560S remain without treatment options [9].

Objective: Identify novel drug candidates for the rare R560S CFTR mutation.

Hypothesis: Computational docking will identify molecules similar to Trikafta’s components that bind the Alpha-Fold predicted structure of the R560S CFTR mutant with high binding affinity and favourable drug-likeness properties.

How?

1) Generated protein structures

Used AlphaFold3 to predict wildtype, ΔF508, and R560S CFTR from UniProt amino acid sequence.

For ΔF508, removed F at position 508 of sequence; For R560S, replaced R with S at position 560 of sequence.

2) Validated AlphaFold accuracy

Superimposed AF3 wildtype with solved PDB structure (PDB: 5UAK) [11] in ChimeraX.

Recorded root mean square deviation (RMSD).

3) Prepared ligands

Copied SMILES for tezacaftor, ivacaftor, and elexacaftor from PubChem.

Found ZINC IDs and tranches.

Downloaded the first 100 similar molecules per component (300 total) from ZINC20 as SDF files.

4) Validated docking protocol

Docked Trikafta components into solved ΔF508 structure (PDB: 8EIQ) [10] using Autodock Vina on PyRx (Figure 5 with ΔF508 and Trikafta).

Confirmed binding locations.

5) Prepared R560S target

Added polar hydrogens to AF3 R560S in ChimeraX.

Saved file as PDB for docking.

6) Performed docking.

Used Autodock Vina on PyRx to dock all 300 ligands against R560S-CFTR with blind docking (exhaustiveness = 8, Figure 5).

7) Filtered results.

Applied thresholds: binding affinity ≤ -9.9 kcal/mol [12] and RMSD lower bound < 2.0 Å [13].

Saved results in Excel sheet.

8) Analyzed drug-likeness.

Ran SwissADME on compounds that passed the filtering thresholds.

Chose candidates based on Lipinski, Ghose, Veber, Egan, and Muegge rules: maximum of 1 violation per rule for each candidate [14].

9) Visualized interactions.

Used ChimeraX to analyze hydrogen bonds (distance tolerance 0.600 Å), hydrophobic surfaces, and electrostatic surfaces for each top candidate.

What?

AlphaFold Structural Validation

Superposition of the AlphaFold3 wildtype with the solved structure (Figure 6a) confirms structural accuracy for docking. The higher all-pair RMSD reflects unstructured regions irrelevant to docking, though some structural variations may exist, particularly in the nucleotide-binding domain. Superposition of the AF3 wildtype with AF3 ΔF508 (Figure 6b) shows minimal structural differences throughout nearly the entire protein, which may explain the success of drugs for this mutant. Superposition of AF3 wildtype with AF3 R560S (Figure 6c) indicates greater divergence from wildtype than ΔF508, with substantial deviations in the nucleotide-binding domain.

pLDDT confidence metrics (Figure 7) were consistently high in structured regions. Some low-confidence predictions in the R560S regulatory and nucleotide-binding domains warrant caution when interpreting docking results in these regions.

Docking Screen Results

Multiple ligand docking was performed using PyRx AutoDock Vina with 300 Trikafta analogs against AF3 R560S. Filtering thresholds were binding affinity ≤ -9.9 kcal/mol (~100 nM) and RMSD < 2.0 Å. A total of 62 compounds passed.

Elexacaftor analogs: 300 screened → 900 configurations → 16 remained after filtering

Tezacaftor analogs: 300 screened → 896 configurations → 45 remained

Ivacaftor analogs: 300 screened → 342 configurations → 1 remained

SwissADME Drug‑Likeness Analysis

SwissADME analysis of the 62 compounds identified 6 meeting ≤1 violation for each of the Lipinski, Ghose, Veber, Egan, and Muegge drug-likeness rules.

ZINC000001157652 (Ivacaftor analog)

Violations: Lipinski 0, Ghose 1, Veber 0, Egan 0, Muegge 1.

ZINC000000650653 (Elexacaftor analog)