Introduction:

The performance of different gene-finders were evaluated on the accuracy measures defined below. The various measures have been taken as defined in Burset and Guigo, 1996 and Rogic et al, 2001.

• NUCLEOTIDE LEVEL ACCURACY:
  
  
  1. True Positives (TP):

     The number of nucleotide bases that are actually coding and are predicted as coding.

  2. True Negatives (TN):

     The number of nucleotide bases that are actually non-coding and are predicted as non-coding.

  3. False Positives (FP):

     The number of nucleotide bases that are actually non-coding and are predicted as coding.

  4. False Negatives (FN):

     The number of nucleotide bases that are actually coding and are predicted as non-coding.

  5. Sensitivity (Sn):

     .

  6. Specificity (Sp):

     .

  7. Correlation Coefficient (CC):

     .

  8. Simple Matching Coefficient (SMC):

     .

  9. Average Conditional Probability (ACP):

     .

  10. Approximate Correlation (AC):

      .




• EXON LEVEL ACCURACY:
  
  
  1. Sensitivity (ESn):

     .

  2. Specificity (ESp):

     .

  3. Average (EAvg):

     .

  4. Correct Exons (CR):

     Proportion of predicted exons whose both ends are correct.

  5. Partially Correct Exons (PC):

     Proportion of predicted exons whose either 5' or 3' alone is correct.

  6. Overlapping Exons (OL):

     Proportion of predicted exons whose end are incorrectly predicted but overlaps an actual exon.

  7. Missed Exons (ME):

     Proportion of actual exons which do not overlap any of the predicted exons.

  8. Wrong Exons (WE):

     Proportion of predicted exons which do not overlap any of the actual exons.




• PROTEIN LEVEL ACCURACY:
  
  
  1. Sensitivity (Psen):

     Ratio of number of predicted protein-coding genes whose entire protein translation product correctly matches the actual protein product of the corresponding genes against number of actual protein-coding genes.

  2. Specificity (Pspe):

     Ratio of number of protein-coding genes correctly predicted by the programs against total number of predicted protein-coding genes.

New Accuracy Measures:

Two new accuracy measures introduced recently are used here to evaluate the predictions. These are the q⁸ and q⁹. For detailes see reference (Zhang and Zhang, 2002).

• q^8:
  
  
• q^9: