Gene Encoding Measures

For the following sensors, assume: sequence S is a string length l = j - i + 1 composed of nucleotide alphabet A,C,G,T. As codons the sequence is described as: , ... , where .

In log notation, this becomes:

Length
A histogram (binsize=1 codons) (see section ) was built from the distribution of gene lengths in the train set. The probability of a gene given its length is the number of train set genes whose are in the same bin as the test gene divided by the total number of genes in the train set. The histogram was smoothed by averaging each bin with a window of 50 and removing genes which were not believed to be protein-encoding. The average gene was slightly less than 1.0Kb long.

Start Codon
The Start Codon sensor is a vector of length 64 which contains a probability of occurence for each start codon . In reality, only three codons have nonzero probabilities: ATG, GTG and TTG.

 

Gene Codon
The Gene Codon sensor is a vector of length 64 which contains a probability of occurence for each codon . The probability is the number of occurences of that codon within coding regions of the train set divided by the total number of codons in the coding regions of the train set. See section for the scores given by the log probabilities for the sensors that were derived from the training set.

 

Stop Codon
The Stop Codon sensor is a vector of length 64 which contains a probability of occurence for each start codon . In reality, only three codons have nonzero probabilities: TAA, TGA, and TAG.

 

Translational Initiation
Using a collection of 1055 E. coli ribosome binding sites from Barrick et. al. [2] a weight matrix of the 4 nucleotides for the sequence -12 to +2 was built. A sequence logo model [10] of this can be seen in Figure 5a of the Barrick article. The score of a sequence according to the promoter sensor is given as: