Splice dataset
The information is a replica of the notes for the Splice dataset from the
UCI repository of machine learning databases.
1. Title of Database:
Primate splice-junction gene sequences (DNA)
with associated imperfect domain theory
2. Sources:
- Creators:
- all examples taken from Genbank 64.1 (ftp site: genbank.bio.net)
- categories "ei" and "ie" include every "split-gene"
for primates in Genbank 64.1
- non-splice examples taken from sequences known not to include
a splicing site
- Donor: G. Towell, M. Noordewier, and J. Shavlik,
{towell,shavlik}@cs.wisc.edu, noordewi@cs.rutgers.edu
- Date received: 1/1/92
3. Past Usage:
- machine learning:
- M. O. Noordewier and G. G. Towell and J. W. Shavlik, 1991;
"Training Knowledge-Based Neural Networks to Recognize Genes in
DNA Sequences". Advances in Neural Information Processing Systems,
volume 3, Morgan Kaufmann.
- G. G. Towell and J. W. Shavlik and M. W. Craven, 1991;
"Constructive Induction in Knowledge-Based Neural Networks",
In Proceedings of the Eighth International Machine Learning
Workshop, Morgan Kaufmann.
- G. G. Towell, 1991;
"Symbolic Knowledge and Neural Networks: Insertion, Refinement, and
Extraction", PhD Thesis, University of Wisconsin - Madison.
- G. G. Towell and J. W. Shavlik, 1992;
"Interpretation of Artificial Neural Networks: Mapping
Knowledge-based Neural Networks into Rules", In Advances in Neural
Information Processing Systems, volume 4, Morgan Kaufmann.
- attributes predicted: given a position in the middle of a window
60 DNA sequence elements (called "nucleotides" or "base-pairs"),
decide if this is a
a) "intron -> exon" boundary (ie) [These are sometimes called
"donors"]
b) "exon -> intron" boundary (ei) [These are sometimes called
"acceptors"]
c) neither (n)
- Results of study indicated that machine learning techniques (neural
networks, nearest neighbor, contributors' KBANN system) performed as
well/better than classification based on canonical pattern matching
(method used in biological literature).
4. Relevant Information Paragraph:
Splice junctions are points on a DNA sequence at which `superfluous' DNA is
removed during the process of protein creation in higher organisms. The
problem posed in this dataset is to recognize, given a sequence of DNA, the
boundaries between exons (the parts of the DNA sequence retained after
splicing) and introns (the parts of the DNA sequence that are spliced
out). This problem consists of two subtasks: recognizing exon/intron
boundaries (referred to as EI sites), and recognizing intron/exon boundaries
(IE sites). (In the biological community, IE borders are referred to
a ``acceptors'' while EI borders are referred to as ``donors''.)
This dataset has been developed to help evaluate a "hybrid" learning
algorithm (KBANN) that uses examples to inductively refine preexisting
knowledge. Using a "ten-fold cross-validation" methodology on 1000
examples randomly selected from the complete set of 3190, the following
error rates were produced by various ML algorithms (all experiments
run at the Univ of Wisconsin, sometimes with local implementations
of published algorithms).
System | Neither | EI | IE |
KBANN | 4.62 | 7.56 | 8.47 |
BACKPROP | 5.29 | 5.74 | 10.75 |
PEBLS | 6.86 | 8.18 | 7.55 |
PERCEPTRON | 3.99 | 16.32 | 17.41 |
ID3 | 8.84 | 10.58 | 13.99 |
COBWEB | 11.80 | 15.04 | 9.46 |
Near. Neigh. | 31.11 | 11.65 | 9.09 |
Type of domain: non-numeric, nominal (one of A, G, T, C)
5. Number of Instances: 3190
6. Number of Attributes: 61
- class (one of n, ei, ie)
- 60 sequential DNA nucleotide positions
7. Attribute information:
- Statistics for numeric domains: No numeric features used.
- Statistics for non-numeric domains
- Frequencies:
| Neither | EI | IE |
A | 24.984% | 22.153% | 20.577% |
G | 25.653% | 31.415% | 22.383% |
T | 24.273% | 21.771% | 26.445% |
C | 25.077% | 24.561% | 30.588% |
D | 0.001% | -- | 0.002% |
N | 0.010% | 0.010% | -- |
S | -- | -- | 0.002% |
R | -- | -- | 0.002% |
Attribute | Description |
1 | One of {n ei ie}, indicating the class. |
2-61 | The remaining 60 fields are the sequence, starting at
position -30 and ending at position +30. |
8. Missing Attribute Values: none
9. Class Distribution:
EI | 767 (25%) |
IE | 768 (25%) |
Neither | 1655 (50%) |
10. Modifications for Delve
- The name attribute in the original UCI distribution was
deleted.
- Cases having with any of D,N,S and R categories have been deleted because
these categories have very low incidence (see the attribute frequency table
above).