The original ORF prediction from the 2004/2005 genome annotation (AtFH3a, At4g15200) misses some conserved sequences, while related motifs are found in the originally predicted neighboring gene (AtFH3b, At4g15190 - dropped in 2005 annotation), indicating that the two predicted genes are actually parts of one. This suspicion is also supported by the GenScan prediction (see above), which, however, misses an exon predicted by TIGR, and adds three more exons on the 5' terminus of the ORF. (Since these exons overlap with the well-characterized neighboring beta-amylase gene, they were excluded from the revised AtFH3 prediction). A very unlikely "13 bp intron" (blue arrow) is more likely an artifact due to a sequencing error; another error has been inferred at the site marked by green arrow (see Cvrckova et al. 2004 for more detail).
A predicted cDNA sequence, differing from the 2005 annotation, has been therefore assembled (download here).
A long continuous stretch free of ATGs in front of the start codon indicates that the real beginning of the ORF may be still missing.