Combining pattern discovery and discriminant analysis to predict gene co-regulation

MOTIVATION

Several pattern discovery methods have been proposed to detect over-represented motifs in upstream sequences of co-regulated genes, and are for example used to predict cis-acting elements from clusters of co-expressed genes. The clusters to be analyzed are often noisy, containing a mixture of co-regulated and non-co-regulated genes. We propose a method to discriminate co-regulated from non-co-regulated genes on the basis of counts of pattern occurrences in their non-coding sequences.

METHODS

String-based pattern discovery is combined with discriminant analysis to classify genes on the basis of putative regulatory motifs.

RESULTS

The approach is evaluated by comparing the significance of patterns detected in annotated regulons (positive control), random gene selections (negative control) and high-throughput regulons (noisy data) from the yeast Saccharomyces cerevisiae. The classification is evaluated on the annotated regulons, and the robustness and rejection power is assessed with mixtures of co-regulated and random genes.

Intraoperative dynamic dosimetry for prostate implants

This paper describes analytic tools in support of a paradigm shift in brachytherapy treatment planning for prostate cancer--a shift from standard pre-planning to intraoperative planning using dosimetric feedback based on the actual deposited seed positions within the prostate. The method proposed is guided by several desiderata: (a) bringing both planning and evaluation in the operating room (i.e. make post-implant evaluation superfluous) therefore making rectifications--if necessary--still achievable; (b) making planning and implant evaluation consistent by using the same imaging system (ultrasound); and (c) using only equipment commonly found in a hospital operating room. The intraoperative dosimetric evaluation is based on the fusion between ultrasound images and 3D seed coordinates reconstructed from fluoroscopic projections. Automatic seed detection and registration of the fluoroscopic and ultrasound information, two of the three key ingredients needed for the intraoperative dynamic dosimetry optimization (IDDO), are explained in detail. The third one, the reconstruction of 3D coordinates from projections, was reported in a previous article. The algorithms were validated using a custom-designed phantom with non-radioactive (dummy) seeds. Also, fluoroscopic images were taken at the conclusion of an actual permanent prostate implant and compared with data on the same patient obtained from radiographic-based post-implant evaluation. To offset the effect of organ motion the comparison was performed in terms of the proximity function of the two seed distributions. The agreement between the intra- and post-operative seed distributions was excellent.

Towards integrating functional imaging in the treatment of prostate cancer with radiation: the registration of the MR spectroscopy imaging to ultrasound/CT images and its implementation in treatment planning

PURPOSE

Dose-escalation to intraprostatic tumor deposits detected by magnetic resonance spectroscopy (MRS) is an example of tumor-targeted radiation therapy. Because treatment planning for prostate brachytherapy is performed based on ultrasound (US)/computed tomography (CT) images, a sine qua non of this technique is the ability to map MRS-positive volumes (obtained in a gland deformed by the endorectal balloon coil) to the US/CT images. An empirical algorithm designed to perform this function, and its validation, are described.

METHODS AND MATERIALS

Mathematically, the problem of mapping points between the MR and US/CT domains comes to: (a) ascertaining that the position of any point in the interior of the prostate is uniquely determined by the shape of the gland, and (b) finding an algorithm that describes this relationship. The image registration algorithm described here is based on the assumption that points within the gland maintain the same relative position with respect to both the axial contours of the prostate and the center of the prostate along the superior-inferior direction. Relative positions of MRS-positive voxels are calculated with this method in both MR and US/CT space. For a particular voxel in the MR space, one obtains first the z coordinate in the US/CT space, that is, along the superior-inferior direction. This determines the axial slice in the US/CT frame of reference where the other two coordinates (x, y) will be calculated. The validity of this algorithm was examined with the aid of a pelvic phantom built to simulate realistically the prostate and its surrounding bony and tissue structures and with CT scans of implanted patients obtained, at several weeks' intervals, as part of an edema-resolution study. Seventy-five "dummy" seeds were placed in the phantom, within the simulated prostate gland, in a quasi-regular pattern. The coordinates of these seeds were determined and thus served as markers of prostate deformation when an inflated rectal probe was introduced in the phantom. CT images of this phantom were taken for different volumes of the MR rectal probe and in each case the prostate outlines were contoured and seed coordinates calculated. Using these data, the predictions of the mapping algorithm could be directly verified.

RESULTS

Absolute values of the 3D-positional errors in this algorithm were 2.2 mm +/- 1.2 mm (average +/- SD). Only 6 of 75 seeds had positional displacement of 4 mm or more. Similar results were obtained in the patient analysis.

CONCLUSIONS

In comparison to the MRS voxel size (6.25 x 6.25 x 3.0 mm3), the present algorithm achieves the desired clinical accuracy. As well, with this 3D algorithm seed positions are reconstructed with an uncertainty that, along the z direction, is less than half the thickness of the typical US slice (0.5 cm).

Operator-free, film-based 3D seed reconstruction in brachytherapy

In brachytherapy implants, the accuracy of dose calculation depends on the ability to localize radioactive sources correctly. If performed manually using planar images, this is a time-consuming and often error-prone process-primarily because each seed must be identified on (at least) two films. In principle, three films should allow automatic seed identification and position reconstruction; however, practical implementation of the numerous algorithms proposed so far appears to have only limited reliability. The motivation behind this work is to create a fast and reliable system for real-time implant evaluation using digital planar images obtained from radiotherapy simulators, or mobile x-ray/fluoroscopy systems. We have developed algorithms and code for 3D seed coordinate reconstruction. The input consists of projections of seed positions in each of three isocentric images taken at arbitrary angles. The method proposed here consists of a set of heuristic rules (in a sense, a learning algorithm) that attempts to minimize seed misclassifications. In the clinic, this means that the system must be impervious to errors resulting from patient motion as well as from finite tolerances accepted in equipment settings. The software program was tested with simulated data, a pelvic phantom and patient data. One hundred and twenty permanent prostate implants were examined (105 125I and 15 103Pd) with the number of seeds ranging from 35 to 138 (average 79). The mean distance between actual and reconstructed seed positions is in the range 0.03-0.11 cm. On a Pentium III computer at 600 MHz the reconstruction process takes 10-30 s. The total number of seeds is independently validated. The process is robust and able to account for errors introduced in the clinic.

The Anderson nomograms for permanent interstitial prostate implants: a briefing for practitioners

PURPOSE

The objective of this report is to re-evaluate the role of the Anderson nomograms in treatment planning for permanent prostate implants. The incentive for revisiting this topic concerns three issues: (1) Although nomograms continue to be used in many centers for ordering seeds, few centers use them during treatment planning; (2) Whereas nomograms were designed to deliver a minimum peripheral dose for a uniform distribution of seeds in the gland, many practitioners use peripheral seed loading patterns to reduce urethral toxicity; and (3) As preoperative and intraoperative treatment planning is becoming standard, the apparent role of nomograms is diminished. The nomogram method is reviewed in terms of: (1) total activity predicted, (2) target coverage (as planned in the operating room and as calculated from postimplant computed tomography studies), and (3) reproducibility (i.e., patient-to-patient and planner-to-planner variability). In each case, the computer-optimization system for intraoperative planning currently in use at our institution was taken as the "gold standard."

METHODS AND MATERIALS

We compared for the same patient the results of nomogram planning to those yielded by genetic algorithm (GA) optimization in terms of total activity predicted (n = 20 cases) and percent target coverage (n = 5 cases). Furthermore, we examined retrospectively the dosimetry of 61 prostate implants planned with the GA (n = 27) and the current implementation of Anderson nomograms (n = 34).

RESULTS

Nomogram predictions of the total activity required are in good agreement (within 10%) with the GA-planned activity. However, computer-optimized plans consistently yield superior plans, as reflected in both pre- and postimplant analyses. We find also that user (specifically, treatment planner) implementation of the nomograms may be a major source of variability in nomogram planning-a difficulty to which robust computer optimization is less prone.

CONCLUSIONS

Nomograms continue to be useful tools for predicting the total required activity for volume implants, and thus for performing an independent check of this quantity. Not unexpectedly, computer optimization remains the preferred planning method. Generally, nomogram-guided implants do not incorporate structures other than the treatment volume into the planning process. Further yet, they deliver a lower dose than that prescribed and result in greater variability among plans than computer-optimized treatments. In summary, nomograms (1) remain an efficient quality assurance tool for computer-generated plans, (2) serve as a good predictor of the number of seeds required for ordering purposes, and (3) provide a simple and dependable backup planning method in case the intraoperative planning system fails.

An independent dose-to-point calculation program for the verification of high-dose-rate brachytherapy treatment planning

PURPOSE

We describe computer software that performs, quickly and accurately, secondary dose calculations for high-dose-rate (HDR) treatment plans, including those employed for prostate treatments.

METHODS

The program takes as primary input the data file used by the HDR remote afterloader console for treatment. Dosimetric calculations are performed using the Meisberger polynomial and the anisotropy table for the HDR Iridium-192 source. For standard applicators, treatment geometry is automatically reconstructed and the dose is calculated at relevant reference point(s). Template-based treatment plans (e.g., prostate) require additional user input; the dose calculation is then performed at user-selected reference points. A total dwell time calculation for volume and planar implants using the Manchester tables was also implemented.

RESULTS

For fixed-geometry HDR procedures, secondary dose calculations are within 2% of the treatment plan, and results are available for review instantly. For more general applications, the calculated and planned doses are typically within 3% at the prescription isodose line. The Manchester-based dwell time calculation is within 10% of the planned time.

Directed evolution of biosynthetic pathways. Recruitment of cysteine thioethers for constructing the cell wall of Escherichia coli

We report that expansion of thioether biosynthesis in Escherichia coli generates sulfur-containing amino acids that can replace meso-diaminopimelate, the essential amino acid used for cross-linking the cell wall. This was accomplished by jointly overexpressing the metB gene coding for L-cystathionine gamma-synthase and disrupting the metC gene, whose product, L-cystathionine beta-lyase, is responsible for the destruction of L-cystathionine and other L-cysteine thioethers. As a result, meso-lanthionine and L-allo-cystathionine were produced endogenously and incorporated in the peptidoglycan, thereby enabling E. coli strains auxotrophic for diaminopimelate to grow in its absence. Thus, current techniques of metabolic engineering can be applied to evolving the chemical constitution of living cells beyond its present state.

The 5' region of Tacaribe virus L RNA encodes a protein with a potential metal binding domain

We have just completed the Tacaribe arenavirus (TV) genome structure by sequencing the 5' region of the L RNA. Analysis of the sequence has indicated the existence of an open reading frame (ORF) in the viral sense RNA encoding a 95 amino acid polypeptide. The first in phase AUG codon is in positions 70-72 from the 5' end of the viral RNA surrounded by a sequence favorable for the initiation of protein synthesis. The ORF ends at positions 355-357. The predicted polypeptide (P11) contains a cysteine-rich sequence bearing a remarkable similarity to the "zinc finger" sequences found in a number of proteins. We have recently reported that the 3' region of the TV L RNA encodes a polypeptide comprising 2210 amino acids in the viral-complementary sequence. This latter gene, i.e., the L gene, terminates at positions 442-440 from the 5' end of the viral RNA. The two genes encoded by the L RNA (L and P11) are in opposite strands of the RNA in sequences that do not overlap, but are separated by a noncoding intergenic region of 82 nucleotides. The nucleotide sequence of the intergenic region leads to the prediction of a strong secondary structure.

Cooperative tandem binding of met repressor of Escherichia coli

We present biochemical and genetic data to support the hypothesis that the Escherichia coli met repressor, MetJ, binds to synthetic and natural operator sequences in tandem arrays such that repression depends not only on the affinity of the DNA-protein interaction, but also on protein-protein contacts along the tandem array. This represents a novel form of regulatory switch. Furthermore, there seems to be homology between the organization of the met and trp operators.

Pressure-induced changes in the secondary structure of the Escherichia coli methionine repressor protein

The effect of hydrostatic pressure on the conformational properties of the E. coli methionine repressor protein in aqueous solution was investigated by infrared spectroscopy. Changes in hydrostatic pressure produce dramatic changes in the spectral region of the conformation-sensitive amide I band. As the pressure is raised up to 18 kbar, the protein undergoes a rearrangement of alpha-helical segments into beta-type structures; after the pressure is released the beta-strands reconvert into less ordered alpha-helical or random segments.

Cell type-specific expression of the human transferrin gene. Role of promoter, negative, and enhancer elements

Transferrin (Tf), the iron transport protein of vertebrate serum, is mainly synthesized in the liver. cis-Acting DNA elements required for liver-specific expression of the human Tf gene were identified by transient and stable expression assays in human hepatoma (HepG2 and Hep3B) and epithelial carcinoma (HeLa) cell lines. Deletion analysis of the 5' DNA sequences of the gene have defined four functionally different regions: (a) A cell type-specific promoter located between positions -125 and -45 which interacts with two nuclear factors and is sufficient for liver-specific expression. (b) A distal promoter region from -620 to -125 base pairs containing positive and negative cis-acting elements which regulate the promoter activity. (c) A negative-acting region between -1.0 and -0.6 kilobase pairs which down-regulates transcription from the Tf promoter. (d) An enhancer located between -4.0 and -3.3 kilobase pairs which is more active in hepatoma than in HeLa cells. Thus, Tf gene expression is modulated by a combination of multiple positive and negative cis-acting elements. The expression results are discussed with respect to our previous description of the trans-acting factors interacting with the proximal and distal promoter regions.

Tacaribe virus L gene encodes a protein of 2210 amino acid residues

The nucleotide sequence of Tacaribe virus (TV) L gene was obtained from two sets of overlapping cDNA clones constructed by walking along the virus L RNA using two successive synthetic DNA primers. Analysis of the sequence indicated the existence of a unique long open reading frame in the viral complementary strand. The first in-phase AUG codon is in positions 31-33 from the 5' end of the viral complementary L RNA surrounded by a sequence favorable for initiation of protein synthesis. The open reading frame ends at positions 6661-6663. The predicted TV L protein is a 2210 amino acid long polypeptide with an estimated molecular weight of 251,942. Comparison of the amino acid sequence of TV L protein with peptide sequences predicted from L-derived cDNA clones of lymphocytic choriomeningitis virus shows an overall 42% of homology.

Interaction of DNA-binding proteins with the tissue-specific human apolipoprotein-AII enhancer

The identification of the binding sites for liver nuclear proteins present in the enhancer that control the cell specific transcription of the human apolipoprotein AII gene is reported. Five adjacent binding sites (motifs I to V) were identified. The motifs III, IV and V can be occupied differently by liver or HeLa nuclear proteins. Two hypersensitive zones (between motifs II-III and IV-V) are present only when liver nuclear extracts were tested. A first characterization of the factors reveal that motif IV interacts with the hepatic transcription factors Tf-LF1 (29) and LF-A1 (28, 30). A CCAAT binding protein, different from CTF/NF1, appears to bind to the motif II. The different binding sites share specific DNA sequences principally with 5' regulatory regions of other apolipoprotein genes.

Different liver nuclear proteins binds to similar DNA sequences in the 5' flanking regions of three hepatic genes

The proximal promoter region of the human transferrin gene contains an hepatocyte-specific cis-element (PRI, nucleotides -76 to -51) whose DNA sequence is homologous to a sequence (nucleotides -89 to -68) present in the transcriptionally essential 5' region of the human antithrombin III gene and to another hepatocyte-specific sequence (A domain) of the human alpha 1-antitrypsin gene promoter. The results reported here lead to the conclusion that the liver trans-acting factor Tf-LF1, binding to the transferrin PRI cis-element interacts with the homologous antithrombin III region, but is different from the transcription factor LF-A1 interacting with the A domain of the alpha 1-antitrypsin promoter. The distal region DRI (nucleotides -480 to -454) of the human transferrin gene promoter presents in its core the same 10 nucleotide-long sequence as the PRI cis-element. We have previously shown that the liver protein Tf-LF2, binding to the DRI element is different from the Tf-LF1 trans-acting factor. In this paper we also show that Tf-LF2 is different from the transcription factor LF-A1 interacting with the alpha 1-antitrypsin promoter. The results allow us to conclude that at least three distinct liver nuclear proteins bind to different subsets of 5' DNA regions containing similar sequences. These sequences are present in genes expressed essentially in liver.