Help about rank differential gene

[BourbonZhang]

Dear SUPPA2 Development Team,

I am researching the differences in alternative splicing of certain genes among different subtypes of breast cancer, hence I need to filter out genes with significant differences across various subtypes.

After conducting differential analysis of evnet and isoform (diffSplice), I obtained their respective .dpsi files, which contain many genes with p-values less than 0.05. How should I rank the results of events and isoforms to obtain a gene ranking?

Best regards and keep up the good work, Bourbon

[EduEyras]

Hi,

Thanks for your question.

If I understand correctly, you want to select those cases that are likely more relevant in separating the subtypes.

You could do something analogous to what one does with differential gene expression when obtaining a ranking to use for e.g. gene set enrichment analysis.

You could use the value deltaPSI ( - 1) log10(p-value)

This will give you a rank with the extreme positive values being the more significant increases in PSI and extreme negative values being the more significant reductions in PSI.

Alternatively you can use |deltaPSI| (-1) log10(p-value) This will rank by significant change regardless of the direction.

Was something like this what you had in mind?

best

Eduardo

On Tue, 2 Apr 2024 at 21:11, bourbonzhang @.***> wrote:

Dear SUPPA2 Development Team,

I am researching the differences in alternative splicing of certain genes among different subtypes of breast cancer, hence I need to filter out genes with significant differences across various subtypes.

After conducting differential analysis of evnet and isoform (diffSplice), I obtained their respective .dpsi files, which contain many genes with p-values less than 0.05. How should I rank the results of events and isoforms to obtain a gene ranking?

Best regards and keep up the good work, Bourbon

— Reply to this email directly, view it on GitHub https://github.com/comprna/SUPPA/issues/187, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADCZKB7REPR4RYO4MWR5RI3Y3J76HAVCNFSM6AAAAABFTCSMYCVHI2DSMVQWIX3LMV43ASLTON2WKOZSGIZDAMJQGQYTGNY . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[BourbonZhang]

Hi Eduardo,

Thank you very much for your answer; it indeed covers one aspect of what I wanted to know.

An even more important point I want to know is that a gene has some events psi and isoforms psi. The previous method you told me can only select one of them. I want to know, how can I combine both to review the status of a gene comprehensively?

For example,

for gene A, event1 has a value of |deltaPSI| (-1) log10(p-value) = 5, event2 = 4, isoform1 = 3, isoform2 = 2, soform3 = 1.

For gene B, event1 = 2, event2 = 3, isoform1 = 4, isoform2 = 6.

How should I compare these?

best， Bourbon

[EduEyras]

Hi

I guess you want to have a single number per gene that reflects how much of an overall significant splicing change that gene has.

You would need to calculate some summary value per gene based on all the events and/or isoforms.

Since the number of events (and/or isoforms) per gene is variable, you may need to take that into account to avoid the bias from genes with many (or too few) events. This could be done by simply normalising by the number of events (and/or isoforms).

You would need to test how such value behaves, to confirm that is meaningful. For instance, it may make more sense to use only with positive values (i.e. |deltaPSI|*(-log10(p-val)) but you could try too with just |deltaPSI| )

Also, |deltaPSI|(-log10(p-val)) is unbounded, since -log10(p-val) can be arbitrarily large. Unlike |deltaPSI|, which is in [0.1]. Although you may transform |deltaPSI|(-log10(p-val)) values to a bounded distribution using the arctan function.

It would be useful to plot the distribution of these values for genes with or without significant events (and/or isoforms) to confirm that such value provides a good separation.

These are just a few ideas of how I would start looking at this.

I hope this helps

Eduardo

On Tue, 2 Apr 2024 at 23:59, bourbonzhang @.***> wrote:

Hi Eduardo,

Thank you very much for your answer; it indeed covers one aspect of what I wanted to know.

An even more important point I want to know is that a gene has some events psi and isoforms psi. The previous method you told me can only select one of them. I want to know, how can I combine both to review the status of a gene comprehensively?

For example,

for gene A, event1 has a value of |deltaPSI| (-1) log10(p-value) = 5, event2 = 4, isoform1 = 3, isoform2 = 2, soform3 = 1.

For gene B, event1 = 2, event2 = 3, isoform1 = 4, isoform2 = 6.

How should I compare these?

best， Bourbon

— Reply to this email directly, view it on GitHub https://github.com/comprna/SUPPA/issues/187#issuecomment-2031976959, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADCZKB6TVK2H65Y6HXPVJH3Y3KTRHAVCNFSM6AAAAABFTCSMYCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMZRHE3TMOJVHE . You are receiving this because you commented.Message ID: @.***>

[BourbonZhang]

Hi Eduardo,

Thank you very much for your answer, these contents are very helpful to me.

best, Bourbon