For the first question I'm not entirely sure I follow what you mean. If you mean if the gene is on the negative strand, should you reverse complement the genomic sequence to get the transcript sequence, then yes. However, I'd generally recommend using a tool like gffread or gff2fasta to ge the sequence of splice isoforms. To look for stop codons, you need to translate the sequence in the correct reading frame and look for codons that correspond to STOP codons. Usually, but not absolutely always, the correct reading frame is the frame of the first ATG in the transcript sequence. I think there are tools that can help you with this. For example, I think IsoformSwitchAnalyszeR can help with these analyses.

In the second question, I would usually recommend mRNA rather than Total RNA. This is because Total RNA contains a lot of immature and nascent RNA where transcription, splicing and other processing has not yet finished.

If I see my gene in the antisense strand 3'-5', should I convert it back to the 5' - 3' direction before identifying any nearby STOP codon in case of a frameshift due to a deletion. (I would like to understand how you would analyze an RNA seq data in terms of identifying any nearby STOP codon during frameshift for both gene in sense and antisense strands).

This question is difficult to answer because it's not clear if you are trying to reinvent some process or simply understand how standard practice works.

In paired-end unstranded RNA-Seq experiments you will see all four types of alignments because of the cDNA process used to amplify RNA and of course the fact there are genes on both genomic strands.

• forward and reverse-complement (single-end and paired)
• reverse, and complement (paired only)

Most people don't go around looking at raw sequences to do this stuff. These different alignment events just wind up a CIGAR strings in your BAM alignment. Splice prediction tools see spliced alignments and gene models and derive events (exon-level, PSI) or entire transcripts from the pileups.

Frameshift effect prediction tools are not concerned with the read you've obtained itself - just the indel in the VCF. Tools like VEP will see a frameshift and predict an early downstream STOP codon.

 CSQ=T|STOP_GAINED|HIGH|TP53|ENSG00000141510|ENST00000269305|7/11|c.300C>T|p.Gln100*|400|300|100|Q/*|CAA/TAA|... 

There may be tools like TransDecoder more concerned with ORF prediction that look at individual sequences. I would imagine they would work better with long reads. I don't know much about them.

So the splicing tools don't really know about frameshifts and the variant effect tools don't really know about splicing.

For aberrant splicing and expression-based analysis what kind of data is best suitable mRNA or Total RNA

poly-A enriched transcripts are actually spliced - that is what you want

total RNA will have a bunch of transcripts with introns in them that aren't in the final product. you won't be able to know if they are retained introns or artefacts.