From Genes to Drugs: overview of computational biology approaches aimed at selecting targets

1
From Genes to Drugs overview of computational
biology approaches aimed at selecting targets
http//www.cse.uconn.edu/cse_sem-conf-instance.php
?id43
Manuel X. Duval PFE/PGRD/GRON/CompSci CSE Dept
Colloquium Oct. 26th, 2007
2

Judgement day
8-12 years.
Discovery
Screen
Lead
Preclinical Dev.
Clinical Dev.
3
Applying Genomics Research into the drug RD
process
Discovery
Development
Post Approval
Regulatory Requests
Disease Genetics
Target Variability
Trial Design
Improving Decision Making, Predicting Efficacy
and Safety
Understanding the Disease
Understanding the Targets
Generating data for regulatory agencies
.
4
Our conceptual framework for rational drug
discovery
One gene
One protein
One small molecule
5

• The 2 leading causes of attrition are
• Toxicity (phase I clinical trial failures)
• Insufficient therapeutic effects (phase II or III
  clinical trial failures)

6
What is targetability ?
7
By targetability, we mean the quality of a gene
to trigger a beneficial measurable biological
outcome when its protein product is altered by
any kind of therapeutic agent
8
Druggability
only and only if one element of the gene family
it belongs to has been shown to be a drug target
The set CHE_REL refers to chemically relevant,
which includes any gene whose product can be
modulated by a therapeutic agent
CHEMICALLY RELEVANT GENOME
9
Targetability
The set BIO_REL refers to biologically relevant,
which includes any gene whose products activity
significantly contributes to a biological outcome
PHYSIOLOGICALLY RELEVANT GENOME
10
Attrition_free_gene (AFG) set
AFG
Whole Genome
CHEMICALLY RELEVANT GENOME DRUGGABILITY
PHYSIOLOGICALLY RELEVANT GENOME TARGETABILITY
11
How could we measure targetability ? Is this a
categorical data type with only 2 values S or
F? Is it rather a continuous random variable ?
PHYSIOLOGICALLY RELEVANT GENOME
12
Have Human genes been created equally targetable?
13
Human polymorphism data is likely to hold the key
to this problem of assigning targetability to
genes.
14
(No Transcript)
15
How to spot the Achilles' heel of the robust
network of the biological pathway universe?

• An Achilles heel is a fatal weakness in spite
  of overall strength, actually or potentially
  leading to downfall. While the mythological
  origin refers to a physical vulnerability,
  metaphorical references to other attributes or
  qualities that can lead to their downfall are
  common.

16
Is there a way to sort genes by their likelihood
to be biologically relevant as drug targets ?
Could we predict blockbuster target genes? What
makes us more alike than different from each
other?
17
The answer is ?
Is polymorphism evenly distributed in the Human
genome? Are the Hs genes equally variable?
no
18
How to assess variability between genes ?
Absolute number of SNPs? Ratio SNP/gene length ?
19
The normalized number of variants estimation of
the nucleotide diversity
It corrects for differing sequence lengths and
sample size (number of chromosomes in the sample
set)
K is the observed number of variants n the
number of chromosomes studied L the sequence
length
20
DNA resequencing data set
coverage sequenced/gene region Polymorphic
sites Non synonymous SNPs Data with respect
to various ethnic populations
Teta_hat.pl --dir --n --file.o
21
Human genes variability distribution
The variable was computed for the whole locus,
for the promoter region, for the exon. For exon,
two variables were derived according to the
non-synonymously outcome of the mutation

The range of the data set is 2740 for
22
Distribution of Log(TetaHatLocus) for 325 loci
23
Among this sample set of 325 genes, which ones
are drug targets?
TARGET_BOOK
dbiOraclegenedb.pfizer.com SELECT gra_biotag,
gra_desc FROM tb_target, tb_mechanism, WHERE
tb_target.gra.idtb_mechanism.mlu_gra_id AND
gra_biotaggb_id
140/325
24
Distribution of Log(TetaHatLocus)
Targetable vs non reported as such
Current drug target in blue
25
Parameters of the normalized number of variants
derived from computing the variables from 325 Hs
loci (Values for the whole gene set and broken
down with respect to the Druggable attribute
according to Target Web values)
26
Same tendency for the genes qualified as
druggable their normalized number of variant
distributions are systematically lower than for
the other gene sets It is likely that our data
set might still be biased towards genes which are
not assigned as targetable only because no
development has been done for them we need a
real training set of genes that failed on Phase
II and III A more detailed statistical test
(logistic regression) suggests that, based on our
data set, there seems to be a correlation
(P0.0099) between targetability and a lower
normalized number of variants value
27

q continuous random variable
Target/non-target categorical variable, with
only two possible outcomes
Logistic regression allows one to assess the
correlation between a continuous random variable
and a dichotomous variable.
28
(No Transcript)
29
Distribution of Log(TetaHatLocus) Kinase vs
other gene families
Kinase are in blue
30
Target (Y1) vs Not target (Y0) P(Yi1)pi P(Yi
0)1-pi E(Yi) pi The logit transformation
links pi with covariates Xi ( )
gsun374/mysql/query-genotypes.pl
--exptNamePFZ_03 --chromosome1 \ --startpos10M
--endpos20M
31
Team Poulabi Banerjee Albert Seymour Michael
Miller
Reference
1. Cargill M, Altshuler D, Ireland J, Sklar P,
Ardlie K, Patil N, Shaw N, Lane CR, Lim EP,
Kalyanaraman N, Nemesh J, Ziaugra L, Friedland L,
Rolfe A, Warrington J, Lipshutz R, Daley GQ,
Lander ES. Characterization of single nucleotide
polymorphisms in coding regions of human genes.
Nat Genet 1999 Nov23(3)373 2. Pungliya MS,
Salisbury BA, Nandabalan K, Stephens JC. Genetic
variability and evolution of two
pharmacologically important classes of
genes. Pharmacogenomics. 2004 Jan5(1)115-27.