some transcripts quantification brief comprehensions

Posted 2020-11-20 beckygogogo

 

 

Some biases in the standard rnaseq analysis

 

 有参拼接：Stringtie 、 Cufflinks and Traph

flow network algorithm ： maximal and minimal methods respectively

 无参组装：Trinity

clustering by K-mers

salmon： as for now the best performance software 

 

DAG：有向无环图 

bagging是减少variance，而boosting是减少bias

 

High variance 是model过于复杂overfit，记住太多细节noise，受outlier影响很大；high bias是underfit，model过于简单，cost function不够好。
A－ bagging随机选取data的subset，outlier因为比例比较低，参与model training的几率也比较低，所以bagging降低了outliers和noise对model的影响，所以降低了variance。
B－boosting参zh Bright的答案，minimize loss function by definition minimize bias.

==========================================================

==========================================================
Streaming fragment assignment for real-time analysis of sequencing experiments

流形碎片的实时测序实验

TIPS：

在估计丰度时候很容易用错或者是正确使用与否是很关键的一步：

RPKM：Reads Per Kilobase of exon modelper Million mapped reads (每千个碱基的转录每百万映射读取的reads)，主要用来对单端测序（single-end RNA-seq）进行定量的方法。
RPKM= total exon reads/ (mapped reads (Millions) * exon length(KB))

 

FPKM：

Fragments Per Kilobase of exon model per Million mapped fragments(每千个碱基的转录每百万映射读取的fragments)，主要是针对pair-end测序表达量进行计算

 

TPM：

Transcripts Per Kilobase of exonmodel per Million mapped reads (每千个碱基的转录每百万映射读取的Transcripts)，优化的RPKM计算方法，可以用于同一物种不同组织的比较。
TPM (推荐软件，RSEM) 的计算公式：

TPMi={( Ni/Li )*1000000 } / sum( Ni/Li+……..+ Nm/Lm )

 

CPM／RPM：

Reads/Counts of exon model per Million mapped reads (每百万映射读取的reads).
RPM的计算公式：
RPM=total exon reads / mapped reads (Millions)