Progress report Yiming Zhang 02/10/2012

All AS events in ASIP Intron retention Exon skipping Alternative Acceptor site NAGNAG AltA Alternative Donor site GYNGYN AltD Alternative both sites (AltP)

NAGNAG alternative splicing Figure 1. NAGNAG alternative splicing with E and I sites and isoforms. NAGNAG alternaive splicing can result in one of three possibilities (Figure 1) - constitutive use of the first acceptor (the so-called exonic, or “E” variant), constitutive use of the second acceptor (the so-called intronic, or “I” variant), or use of both acceptors, that is,alternative splicing (the “EI” variant). Sinha et al. 2010

GYNGYN alternative splicing Figure 2. GYNGYN alternative splicing with e and i sites and isoforms. Hilller et al. 2006

All introns ConstitutiveNAGNAG-ENAGNAG-IGYNGYN-eGYNGYN-I……AlternativeIntronRExonSAltANAGNAG-ei……AltDGYNGYN-ei……Multiple AS……Unclear

Intron statistics from ASIP ATBDGMLJMTOSPPPTSBSLVVTotal Cons.EST>=4all NAG-E NAG-I GYN-e GYN-i EST>=10all NAG-E NAG-I GYN-e GYN-i Alt.EST>=2IntronR AltA_all AltA_NAG AltD_all ALtD_GYN AltP ExonS Table 1. Intron statistics from ASIP. 4 species which have small amount of data are not listed here. All statistics are intron-based instead of event-based which means redundancy has been removed. The most common type of alternative intron type is IntronR, second common type is ExonS. NAGNAG AS occurs much more frequently in AltA than GYNGYN AS occurs in AltD.

Background NAGNAG alternative splicing which can insert or delete a single amino acid in the protein, is very common and well studied in animals. The NAGNAG motif is present in 30% of human genes and is functional in at least 5% of the genes. Hiller et al NAGNAG AS is frame-preserving, the vast majority of cases should lead to different proteins. Studies so far have found evidence of both cases where such proteins have variations in function, as well as those in which there is no noticeable difference. Akerman et al Iida et al The GO analyses in some studies shows genes with specific GO term DNA binding to be statistically significant and more than half of all AS-NAGNAG events affected polar amino acid residues. Iida et al Sinha et al. 2010

Background The studies of NAGNAG AS in plant is few right now (Only 3 species: Arabidopsis, Rice and Physcomitrala). One study found 321 and 372 AS-NAGNAG events in Arabidopsis and rice, respectively. Another study found 6% of all introns and 21% of all annotated genes in Arabidopsis harbor a genomic NAGNAG acceptor motif. Iida et al Schindler et al In addition, the GO analysis is agreed with previous study in human that the specific GO term DNA binding is statistically significant. Some study indicates that NAGNAG acceptors frequently occur in the Arabidopsis genome and are particularly prevalent in SR and SR-related protein-coding genes. Sinha et al 2010 Schindler et al. 2008

Background The state-of-the-art in silico studies for prediction of NAGNAG splice site are done by Sinha's group for both human and plant species. They achieved high balanced specificity and sensitivity for both human and plant species. The most informative features they found are the nucleotides in the NAGNAG and in its immediate vicinity, along with the splice sites scores. The model they trained on human data also can achieve high AUC on plant data shows that NAGNAG splicing in plants is similar to that in animals. Sinha et al. 2009, 2010

NAGNAG dataset I tried to predict NAGNAG events (thus to predict EI, I or E isoforms) based on the dataset I generated from ASIP using Random Forest. Strict criteria has been used to identify NAGNAG events from ASIP database: For E and I events, at least 10 ESTs or cDNAs support them, and for EI events at lease 2 EST or cDNA support each isoform. After removing redundancy, I got 458 EI form alternative NAGNAG introns, 1988 E form constitutive introns and 685 I form constitutive introns in 15 plant species.

Features Figure 3. A total of 28 features which each represented a nucleotide, and thus had four possible values (A, C, G, T). U1, U2, U3 are the first three nucleotides in the upstream exon. D1, D2, D3 are the first three nucleotides in the downstream exon. A weak polypyrimidine tract (PPT) can contribute to AS. So P1-P20 are PPT upstream of NAGNAG. Finally, I also use intron length as an additional feature.

Classifier evaluation Random Forest with 200 trees has been used and 5 fold cross validation has been applied. TP rateFP ratePrecisionRecallF-measureROC areaClass E I EI The evaluation results strongly agree with Sinha’s paper (For Physcomitrella) in which AUC = 0.96, 0.99 and 0.98 for the EI, E and I forms, respectively.

Figure 4. The EI class, or AS, harder to predict (AUC = 0.967) than the two constitutive variants, E and I (AUC = for both).

Most informative features Figure 5. Most informative features according to information gain.

Sequence Logos Figure 6a. Figure 6b.

Figure 6c. Figure 6d. Figure 6a-6d. Sequence logos of NAGNAG splice sites. 6a: E sites; 6b: I sites; 6c: EI sites; 6d: all splice sites. Position 1-3 is U1-U3. Position 4-24 are P20-P1. Position are D1-D3.

Conclusion NAGNAG-AS can be predicted with high accuracy. Using carefully constructed training and test datasets, an in silico performance of AUC = 0.967, and was achieved for the EI, E and I forms, respectively. The most informative features are the nucleotides in the NAGNAG and in its immediate vicinity. NAGNAG AS in plants is similar to that in animals and is largely dependent on the splice site and its immediate neighborhood.