I am currently working on identifying a type of bioactive peptide for which there is no database and I found them by reading articles. The number of peptides that possess my intended characteristic and have been published in valid articles, is about 40 to 50. So the number of positive data is very small. Since we do not have negative reports in biology, I do not know how to find or create negative data.
So two questions are posed:
1. How to build a reliable statistical or machine learning model to identify bioactive peptides using this small set of positive samples?
2. How to increase the number of positive data and find or create negative data such that the model that is made based on these data will be correct and reliable?
I will appreciate it if you provide me with some advice in these regards.
Try checking this data synthesis technique, if it helps
https://www.researchgate.net/publication/343125982_Early_trends_of_socio-economic_and_health_indicators_influencing_case_fatality_rate_of_COVID-19_pandemic?_sg=SkWlIR8oPygckSRaOo99XomtJC2WiHOb4YHqn3ux0u5wsifxTcW_C44WqTEBpditgN15JN1RWrb_uIHpcZbT6DLPcicAi127xrGlmc2S.2iYxCimnfC9diax7cSykmNVXyd6z6qOjDb5mztB-QEOcpxgu_aj6AgOBA-poX47m0Hhcz0OjQPUKBRfW11158w
Hi Maryam,
I would start by understanding the peptides you have better. What is their size? Do they look similar?
If they are at least 15-20 residues long, this should allow clustering them or looking for related sequences in biological databases.
If they are indeed similar you could consider building a Hidden Markov Model, which captures their range of sequences. In essence, see if your problem may be solved using the following approach: Article A Combined Transmembrane Topology and Signal Peptide Prediction Method
A negatives set (or rather a baseline or decoy set) can be created by generating a large number of random peptides (having sizes comparable to sizes of active peptides). A random sequence is highly unlikely to have a desired biological effect. It is possible that a large set of random peptides would accidentally contain a couple of actives, but this wouldn't undermine its value for developing an ML model.
- Badges
- Science topic
- Similar topics
- Biology