CharMark: Unmasking Cognitive Decline in Everyday Speech

Imagine a future where your casual conversation, describing a picture or chatting with a friend, could reveal the earliest whispers of cognitive change.
CharMark brings that future a step closer. It’s a lightweight Python toolkit that turns plain text transcripts into interpretable “fingerprints” of your speech, helping researchers and clinicians spot subtle signs of dementia before they become unmissable.

Why CharMark?

Traditional speech analyses dive into audio waves or complex neural nets—powerful, but often opaque. CharMark asks a different question:

“At the level of single characters and pauses, how does language flow?”

By building a first-order Markov chain from your transcript, then computing its steady-state probabilities, CharMark captures micro-patterns—like hesitation in pauses or recurring letter sequences—that can serve as early digital biomarkers of cognitive decline.

Key Features

• Clean & Simple: Load your CSV of transcripts, call a few methods, and get back matrices of interpretable features.
  
• Interpretable: Character-level probabilities, Kolmogorov-Smirnov tests, PCA plots, and network visualizations lay bare why two groups differ.
  
• End-to-End Pipeline: From raw text cleaning to unsupervised clustering and Lasso-based validation, everything’s in one place.
  
• Lightweight: No massive datasets or GPUs required—just Python and your transcripts.
  

Quickstart

```bash git clone https://github.com/jkevin2010/charmark-biomarker-discovery.git cd charmark-biomarker-discovery pip install -r requirements.txt

```

Example Notebook

Want to see CharMark in action?
Check out example.ipynb for a step-by-step demo using a toy dataset.

This notebook walks you through:
1. Creating a sample CSV
2. Extracting steady-state features
3. Running PCA, KS tests, and Lasso validation
4. Visualizing the character transition network

Cite Us

If you use CharMark in your research, please cite:

Mekulu K, Aqlan F, Yang H (2025). CharMark: Character-Level Markov Modeling to Detect Linguistic Signs of Dementia. Preprint.
DOI: 10.21203/rs.3.rs-6391300/v1

BibTeX: ```bibtex @misc{Mekulu2025CharMark, author = {Kevin Mekulu and Faisal Aqlan and Hui Yang}, title = {CharMark: Character-Level Markov Modeling to Detect Linguistic Signs of Dementia}, year = {2025}, doi = {10.21203/rs.3.rs-6391300/v1}, url = {https://doi.org/10.21203/rs.3.rs-6391300/v1}, note = {Preprint on Research Square} }