Help appreciated!! We are having Empirical data sets of plant barcodes (matK, ITS and rbcL), and want to compare analytical methodology used for Empirical data sets with the Synthetic data sets.
How can we generate such kind of Synthetic datasets? I have found relevant papers but was not able to understand methodology. I would appreciate if any one could provide detailed procedure or provide any tutorial.
REFERENCE: Supervised DNA barcodes paper:
Article Supervised DNA Barcodes species classification: Analysis, co...
From paper....
Synthetic data:
Real DNA Barcode datasets are simulated with Coalescent package in Mesquite version 2.75 (see the related work [8]). The data are simulated considering time of species divergence and the effective population size (Ne), i.e., the number of individuals in a population (of a species) that are contributing genes to the succeeding generations.
Another similar paper...
Conference Paper Species Identification using DNA Barcode Sequences through S...
Hey,
may I ask for what purpose you need this synthetic dataset and how shall it look like and what did you do with your "real" dataset (methodically)?
Regards,
André
André Marquardt and Samir Mishrif Khalaf thanks for your reply, I am using machine learning classifiers to classify species, where we have multi-sequence (DNA) dataset. Presently we have empirical dataset (Real dataset: consisting of singleton or doubleton species, affecting species resolution potential of classifiers) so if we compare efficiency of classifiers with that of Synthetic dataset (Synthesized with >10 specimens per species), we can find out precision and error rate of those classifiers.
Hey,
so if I get you right, you just want to have a dataset that has been randomly generated using more than 10 species in a not further specified ratio? If I am correct with this assumption, I would recommend to just use the reference sequences of your desired species, randomly generate fragments out of those and then randomly combine them.
If I am wrong, I would be happy if you can clarify at which step I got lost.
Regards,
André
Yeh! André Marquardt with the reference sequence of desired species, randomly generate sequences considering species divergence (not more than 2% divergent otherwise it will be considered as different species) and the effective population size (Ne), i.e., the number of individuals in a population (e.g 10 individuals of a species).
So, yet again just for clarification: You would like to have a dataset, consisting of different samples of a chooseable amount, but the data for each sample may/should have spiked in up to 2% of sequences, that are NOT of this sample but of the same species as the "main" sample? Afterwards you are going to run your trained classifier for each sample and would like to know the species as "label"?
If my summary is correct you coul do something like I recommended:
In silico working with random fragments is comparable to in vitro experiments, I have done this before: http://www.haematologica.org/content/104/2/277.long
To be honest, I am not quite sure how your classifier works an DNA sequences, but i would be happy to read the publication in the end.
Regards,
André
Thanks Andre, could you please clarify point number 3.
How to generate a species dependent dataset that is random sampled with a chooseable amount of randomly spiked in fragments of other reference sequences that are within the same species.
I would appreciate if you could suggest any software with tutorial
Hey,
no problem, you're welcome. I really do not know any software for this purpose and do not think there really is one for your purpose. I would recommend to code it on your own, this is at least the way I always do it.
Regarding point 3:
After you downloaded all reference sequences of all possible species and "samples" you want, you can easily cut our fragments of a specific length from this. In my tool I also linearized the genome, such that I do not have to mind chromosomes anymore. After cutting out for example 100.000.000 reads per species/sample you can randomly select fragments out of this database. If you want, you can also additionally choose to spike in some fragments of other species or samples (other reference genomes), leading to a in silico sample, consisting of your randomly generated fragments.
The whole process, how I did it before using human DNA as sample, is described in the publication I linked.
Hope this clarifies it a bit, feel free to ask if not.
Regards,
André
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics
- Bioinformatic Tools