\documentclass{article}

\usepackage{Sweave}

\begin{document}

\title{Solexa QA report}
\date{\today{}}
\maketitle{}

<<setup, echo=FALSE>>=
options(digits=3)
library(ShortRead)
library(lattice)
@

<<qa-run, echo=FALSE>>=
load("@QA_SAVE_FILE@")
@

\section{Overview}

This document provides a quality assessment of Genome Analyzer
results. The assessment is meant to complement, rather than replace,
quality assessment available from the Genome Analyzer and its
documentation. The narrative interpretation is based on experience of
the package maintainer. It is applicable to results from the `Genome
Analyzer' hardware single-end module, configured to scan 300 tiles per
lane. The `control' results refered to below are from analysis of
$\varphi$X-174 sequence provided by Illumina.

An R script containing the code used in this document can be created
with
<<code,eval=false,echo=TRUE>>=
fl <- system.file("template", "qa_solexa.Rnw", package="ShortRead")
Stangle(fl)
@ 

\section{Run summary}

Read counts. Filtered and aligned read counts are reported relative to
the total number of reads (clusters). Consult Genome Analyzer
documentation for official guidelines. From experience, very good runs
of the Genome Analyzer `control' lane result in 6-8 million reads,
with up to 80\% passing pre-defined filters.
<<read-counts>>=
ShortRead:::.ppnCount(qa[["readCounts"]])
@

Base call frequency over all reads. Base frequencies should accurately
reflect the frequencies of the regions sequenced.
<<base-calls>>=
qa[["baseCalls"]] / rowSums(qa[["baseCalls"]])
@ 

Overall read quality. Lanes with consistently good quality reads have
strong peaks at the right of the panel.
<<read-quality-raw, fig=TRUE>>=
df <- qa[["readQualityScore"]]
print(ShortRead:::.plotReadQuality(df[df$type=="read",]))
@ 

\section{Read distribution}

These curves show how coverage is distributed amongst reads. Ideally,
the cumulative proportion of reads will transition sharply from low
to high. 

Portions to the left of the transition might correspond roughly to
sequencing or sample processing errors, and correspond to reads that
are represented relatively infrequently. 10-15\% of reads in a typical
Genome Analyzer `control' lane fall in this category. 

Portions to the right of the transition represent reads that are
over-represented compared to expectation. These might include
inadvertently sequenced primer or adapter sequences, sequencing or
base calling artifacts (e.g., poly-A reads), or features of the sample
DNA (highly repeated regions) not adequately removed during sample
preparation. About 5\% of Genome Analyzer `control' lane reads fall in
this category.

Broad transitions from low to high cumulative proportion of reads may
reflect sequencing bias or (perhaps intentional) features of sample
preparation resulting in non-uniform coverage. Typically, the transition is about
5 times as wide as expected from uniform sampling across the Genome
Analyzer `control' lane.
<<read-distribution-occurrence, fig=TRUE>>=
df <- qa[["sequenceDistribution"]]
print(ShortRead:::.plotReadOccurrences(df[df$type=="read",], cex=.5))
@ 

Common duplicate reads might provide clues to the source of
over-represented sequences. Some of these reads are filtered by the
alignment algorithms; other duplicate reads migth point to sample
preparation issues.
<<common-duplicate-reads>>=
ShortRead:::.freqSequences(qa, "read")
@ 

Common duplicate reads after filtering
<<common-duplicate-mapped-reads>>=
ShortRead:::.freqSequences(qa, "filtered")
@

\section{Cycle-specific base calls and read quality}
Per-cycle base call should usually be approximately uniform across
cycles. Genome Analyzer `control' lane results often show a deline in
A and increase in T as cycles progress. This is likely an artifact of
the underlying technology.
<<per-cycle-base-call, fig=TRUE>>=
perCycle <- qa[["perCycle"]]
print(ShortRead:::.plotCycleBaseCall(perCycle$baseCall))
@ 

Per-cycle quality score. Reported quality scores are `calibrated',
i.e., incorporating phred-like adjustments following sequence
alignment. These typically decline with cycle, in an accelerating
manner. Abrupt transitions in quality between cycles toward the end of
the read might result when only some of the cycles are used for
alignment: the cycles included in the alignment are calibrated more
effectively than the reads excluded from the alignment.
<<per-cycle-quality, fig=TRUE>>=
print(ShortRead:::.plotCycleQuality(perCycle$quality))
@ 

\section{Tile performance}

Counts per tile. The dashed red line in the following plot indicates the 10\% of tiles with
fewest reads. An approximately uniform 
% FIXME (wh 6 June 2009): do you mean uni-modal?
distribution suggests
consistent read representation in each tile. Distinct separation of
'good' versus poor quality tiles might suggest systematic failure,
e.g., of many tiles in a lane, or excessive variability (e.g., due to
unintended differences in sample DNA concentration) in read number per
lane.
<<counts-histo, fig=TRUE>>=
perTile <- qa[["perTile"]]
readCnt <- perTile[["readCounts"]]
cnts <- readCnt[readCnt$type=="read", "count"]
print(histogram(cnts, breaks=40, xlab="Reads per tile",
                panel=function(x, ...) {
                    panel.abline(v=quantile(x, .1),
                                 col="red", lty=2)
                    panel.histogram(x, ...)
                }, col="white"))
@ 

Spatial counts per tile. Divisions on the color scale are quantized,
so that the range of counts per tile is divided into 10 equal
increments. Parenthetic numbers on the scale represent the break
points of the quantized values. Because the scale is quantized, some
tiles will necessarily have `few' reads and other necessarily `many'
reads. 

Consistent differences in read number per lane will result in some
lanes being primarily one color, other lanes primarily another color.
Genome Analyzer data typically have greatest read counts in the center
column of each lane. There are usually consistent gradients from `top'
to `bottom' of each column.

Low count numbers in the same tile across runs of the same flow cell
may indicate instrumentation issues.
<<read-counts-per-tile, fig=TRUE>>=
print(ShortRead:::.plotTileCounts(readCnt[readCnt$type=="read",]))
@ 
 
Median read quality score per tile. Divisions on the color scale are
quantized, so that the range of average quality scores per tile is
divided into 10 equal increments. Parenthetic numbers on the scale
represent the break points of the quantized values.

Often, quality and count show an inverse relation.
<<read-score-per-tile, fig=TRUE>>=
qscore <- perTile[["medianReadQualityScore"]]
print(ShortRead:::.plotTileQualityScore(qscore[qscore$type=="read",]))
@

\section{Alignment}
Mapped alignment score. Counts measured relative to counts in score
category with maximum representation. Successful alignments will be
reflected in a strong peak to the right of each panel.
<<mapped-alignment-score, fig=TRUE>>=
print(ShortRead:::.plotAlignQuality(qa[["alignQuality"]]))
@ 

\end{document}