question about cufflinks

[liuxq]

For paired-end reads, when mapping to reference genome, the mapping hits will be classified as singleton, with mate-paired and properly paired according to samtools flagstat. How does cufflinks deal with these three kinds of mapping reads for abundance estimation and transcript assembly?
For abundance estimation, cufflinks will estimate F(i) for the distribution of fragment. How does it work for singleton mapping? Only use 100bp for 100bp x 2 to estimate the distribution?

Thanks very much and look forward to your response.