Algorithm of determination of circadian gene expression profiles analysed with DNA microarrays

DNA microarrays allow to simultaneously determine the expression level of thousands of genes. A nycthemeral study must enable to conclude which ones show a circadian rhythm. Two aspects prove this to be quite difficult: firstly, what does "circadian" exactly mean and how to quantify this qualification, and secondly which genes pertain to this definition. Our method, derived from linear optimisation procedures, consists in determining a cost function, depending from magnitudes characterising the notion of circadian rhythm. Given number of genes present on the microarray are known to be expressed rhythmically; their time series are considered as reference series. We have further constructed random series having the same temporal structure as the circadian gene series. We then carried out an optimisation procedure to determine the weighting coefficients in order to obtain a cost function value which orders the time series as follows: the reference series are in the first rows and the random series have low scores. We have tested this method on over 6000 genes expressed in mouse liver. We obtained a circadian gene detection probability of 100% with a false positive rate inferior to 1%.