- To establish a basic computational facility for the
implementation of algorithms for bioinformatics and computational
- To conceive, develop, and implement on hardware three
industrially-relevant BCB solutions as a proof of the concept.
- To develop a set of hardware-oriented structural and/or
computational building “blocks” that will serve as the basic bricks for
implementing further advanced BCB solutions.
- To acquire advanced know-how and specific skills in the
development of BCB hardware.
In the framework of this project, we propose to first develop and
realize hardware accelerators for three industrially-relevant BCB
methods or algorithms. They will be implemented on, at least, one of the
two available hardware platforms : a development board integrating a
Virtex5 SX50T with PCIe 4x communication connection with the host PC and
a multi-FPGA board including seven separate Virtex5 LX50 chips, each
with a PCIe 1x connection with the host PC. The three algorithms are the
- PLINK algorithm : PLINK is a whole genome analysis software
often referred as the Swiss knife tool in this very ﬁeld. Among the many
possibilities offered, we will especially focus on the Single
Nucleotide Polymorphisms (SNPs) association and SNP-epistasis involving
millions of pair-wise simple comparisons of vectors, each containing a
huge amount of values.
- Biomarker-based diagnostic decision modeling : A significant
current trend in disease diagnostic decisions is based on measuring
levels of activity or expression, in the organism, of different
substances, the biomarkers. Such a diagnostic decision often depends on
complex patterns of activity from many biomarkers. However, measuring
these biomarkers is costly and thus, there is a pressure towards finding
the minimal set allowing to efficiently diagnose a specific disease.
Fuzzy CoCo, the proposed modeling approach, is a cooperative
co-evolutionary algorithm which has proved to be able to find
highly-predictive candidate systems. Moreover, in addition to their
predictive value, the discovered fuzzy models, being linguistic
rule-based, also offer an explanation for the possible reasons
underlying the diagnostic decisions made.
- Functional modeling of gene regulatory networks : The
functional modeling of gene regulatory networks (GRN) consists of
iteratively updating the values of multiple gene expression values
within a network of many genes (from hundreds to several thousands) in
order to reach a steady or oscillating state. Every gene expression
value usually depends on the expression values of several or many other
genes in the network, following a more or less complex dynamics. As a
result, this iterative update needs an astounding number of evaluations
and the complexity of the computation increases exponentially with the
number of genes within the network and with the number of different
relations linking their level of expression.
In a second phase, from these implementations we will extract some
salient principles that could be used to generalize the process of the
hardware acceleration of further BCB algorithms. These generalizations
will include data transfer and communication requirements, different
structural patterns and dataflows specific to BCB algorithms, although
with a coars-enough grain to be applicable to many different types of
Finally, based on these principles, we will propose a library of
modules, both software and hardware, that should facilitate the
implementation of hardware accelerators for many BCB applications.