Identifier line of each sequence must start with ">".
We build a model with coding and ncRNAs coming from a large spectrum of species from different reigns:
Danio rerio (zebrafish)
Mus musculus (mouse)
Oryza sativa (rice)
For each sequence are computed two values: the probability to be a non-coding RNA and the probability to be a coding RNA.
A prediction is rejected if the difference between the coding and the non-coding probabilities is lower than the rejection threshold.
The rejection option highlights the transcripts which are between the coding and the ncRNAs.
Rejection threshold is a value between 0 and 1
0 : No rejection
1 : Everything is rejected
By default, the rejection threshold is set to 0.8. This value can be set by the user.
The rejection threshold is application dependent. For certain applications we may want a high rejection threshold in order to have the most reliable predictions but in an exploratory analysis, we may use a smaller rejection threshold in order to keep more predictions even if they are potentially misclassified.
Press CTRL+D on the waiting page to add to your favorites and to access results later.
In that section are displayed the number and percentage of Noncoding, Coding and Rejected sequences.
SOM (Kohonen, 2001) is a neural network that is able to cluster and visualize high dimensional data. By using an unsupervised competitive learning algorithm, SOM is able to produce a map representing the input space.
Each circle represents the Best Matching Unit (BMU): the neuron which gather RNA sequences sharing similar features.
In some cases, it could appear some empty areas that refer to BMU without any predicted sequences in it.
"Label" represents the predicted class of the RNA sequences: Rejected, Noncoding, Coding.
The table is constituted with the following columns:
1. Sequence ID: identifier of the sequence
2. BMU (Best Matching Unit) represents the coordinates of the neuron which belong the sequence (see Self-Organizing map)
3. Label: predicted class (Coding, Noncoding or Rejected)
4. Probability of the sequence to be a coding RNA | p(coding)
5. Probability of the sequence to be a non-coding RNA | p(noncoding)
The prediction table is available as a tab-separated text format via the Download button.
Barplots show the number of sequences according to:
On the left: the probability of the sequence to be a coding RNA | p(coding)
On the right: the probability of the sequence to be a noncoding RNA | p(noncoding)
Colors refer to the label or predicted class: Rejected, Noncoding, Coding.
Two plots showing the profiles of the representatives for each feature are available via the Download buttons.
Each point in the figure represents the value of the representative for a given feature.
Our IRSOM algorithm is an alignment-free method based on three types of features which are sequence bias, ORF statistics and k-mer motifs :
Sequence bias: composed of three features, which are the codon position bias, the frequencies of each nucleotide and the GC frequency. The purpose of the codon position bias is to measure if there is nucleotide position bias in codons.
Open reading frame (ORF): we compute the length and coverage of the maximal ORF that are useful to access the information of the most probable coding sequence of the transcript.
k-mer: the k-mers are all the words of size K that are contained in a string. Here, we select the k-mers of size 3, 4 and 5 and compute their frequencies.