AdmiRePred – A method for predicting abundant miRNAs in exosomes
Authors/Creators
-
1.
Indraprastha Institute of Information Technology Delhi
Description
Title:
AdmiRePred Dataset – Abundant and non‑abundant microRNA (miRNA) sequences in blood exosomes
Description:
Project: AdmiRePred – A method for predicting abundant miRNAs in exosomes
Publication: Arora, A., & Raghava, G.P.S. (2025). AdmiRePred: A method for predicting abundant miRNAs in Exosomes. bioRxiv. https://doi.org/10.1101/2025.03.19.644072
Overview: This dataset accompanies AdmiRePred, a method for predicting miRNAs that are abundantly present in blood exosomes under normal conditions. Exosomes carry miRNAs that reflect the physiological state of parent cells, making them promising non‑invasive biomarkers for liquid biopsy (cancer, cardiovascular, neurodegenerative diseases). Unlike prior binary methods (exosomal vs. non‑exosomal), this study uses expression‑based classification to identify miRNAs highly expressed in exosomes, establishing a baseline for disease‑specific variation.
Content: The dataset contains miRNA sequences with expression data from EVmiRNA (blood exosomes, serum/plasma, normal human subjects, n=60) and validated with GEO GSE270497.
| Class | Definition | Count | Length range |
|---|---|---|---|
| Abundant (positive) | Average RPKM > 2 | 348 | 16–25 nt |
| Non‑abundant (negative) | Average RPKM < 1 | 349 | 16–25 nt |
Key Findings:
-
Mann‑Whitney test: 31 features significantly different (p < 0.05) between abundant and non‑abundant miRNAs
-
TF‑IDF features (reverse complement, k‑mer 1,2): C, AC, CG, GA increased in abundant sequences; UU, U, AU, UA decreased
-
Top important features: Binary profiles and TF‑IDF (reverse complement) most discriminative
Best Model Performance (validation set – 140 sequences):
| Model | Features | AUC | MCC | Accuracy |
|---|---|---|---|---|
| Hybrid (ET + BLAST) | Best features + similarity | 0.854 | 0.559 | 77.7% |
| Hybrid (SVC + BLAST) | Best features + similarity | 0.846 | 0.513 | 77.7% |
| ET (ML only) | Best features (binary + TF‑IDF) | 0.769 | 0.410 | 64.0% |
| RF (ML only) | Binary profiles | 0.754 | 0.381 | 69.1% |
| SVC (ML only) | Binary profiles | 0.706 | 0.378 | 68.4% |
Alignment‑based methods – BLAST (e‑value = 10⁻², validation set):
| Metric | Value |
|---|---|
| Correct hits (abundant) | 24 |
| Incorrect hits | 0 |
| Coverage | 24/69 (34.8%) |
BLAST had 0% error but poor coverage; motif‑search had low coverage (8–15 sequences) + high error rate.
Comparison with existing methods (validation set):
| Method | AUC | Notes |
|---|---|---|
| AdmiRePred (hybrid) | 0.854 | Expression‑based |
| EmiRPred | 0.623 | Binary (exosomal vs. non‑exosomal, no expression) |
| miRNALoc | 0.422 | Subcellular localization (exosome one of many locations) |
Data Curation & Quality Control:
-
Source: EVmiRNA (blood exosomes, serum/plasma, normal subjects)
-
Validation: GEO GSE270497 (small RNA sequencing of plasma exosomes, normal subjects)
-
Expression threshold: Abundant = RPKM > 2; Non‑abundant = RPKM < 1
-
Redundancy: Duplicates removed
-
Length range: 16–25 nucleotides
-
Train/validation split: 80/20 (5‑fold CV on training)
Usage: Predicting highly abundant miRNAs in blood exosomes for liquid biopsy biomarker discovery; designing mutant miRNAs with altered exosomal abundance (Design module); similarity search against known abundant/non‑abundant miRNA database (BLAST module); establishing baseline for disease‑specific miRNA variation studies.
Related Resources: Web server: https://webs.iiitd.edu.in/raghava/admirepred/ | GitHub: https://github.com/raghavagps/admirepred
Contact: raghava@iiitd.ac.in (Gajendra P. S. Raghava)
Files
raghavagps/admirepred-v1.0.zip
Files
(669.8 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:db30b4ebe7f908eb9c7fb9f97f12a14c
|
669.8 kB | Preview Download |
Additional details
Related works
- Is supplement to
- Software: https://github.com/raghavagps/admirepred/tree/v1.0 (URL)
Software
- Repository URL
- https://github.com/raghavagps/admirepred