The aim of this wet lab project was to construct a library of RBS parts for L. lactis. With enough variation in the expression of each strain with a particular sequence, we can enable fine-tuning of genes for multiple metabolic engineering goals using this organism.
We also intended to test out computationally generated sequences based on our very own predictor-optimizer model and understand if we are able to predict expression levels reliably.
What we learned from our results is that research is always a mixed bag. We win some and lose some. When we ‘lose’, it indicates that we need to go back to the drawing board, figure out what went wrong, and try again. This cycle of design-build-test-learn happens both at the small scale of day-to-day experiments and at the large scale of testing out new hypotheses. This is illustrated clearly in our lab notebook and our engineering success.
Fluorescence Assay Results
- Fluorescence readings are adjusted for OD and negative control values.
- The sequencing data for each of these sequences is provided below:
- Looking at our results, we have been able to successfully characterize a library of 24 RBS sequences with a good variable range of expression.
- Most of the sequences that were computationally generated were quite low expressing compared to our positive control (PJ102 - natural RBS of L. lactis).
- The strains generated through randomization of the native RBS have shown a large steady range of variation from low to medium to similar expression levels as the positive control.
- One surprise for us was the RBS15 from the computationally generated set that had twice the expression level of the control.
- Future work would include integrating the sequencing data from the randomized strains into the predictor-optimizer model and attempting to predict the expression of the computationally generated sequences using the data from the randomized variants. Characterizing a larger randomized library including sequences that also show higher expression than that of the control will also give us higher tuning ability.
- This would be beneficial to attempt to correct the bias in the original dataset used to train the model and also train the model on L. lactis-specific data to ascertain if better predictions can be made on the computationally generated sequences.
- Our predictor-optimizer model showed good performance on the test set which was similar to the dataset it was trained on. We suspect that due to the dearth of gram-positive bacteria data in the dataset, our model hasn't been able to generalize well to accurately predict expression in Lactococcus lactis.
- This work can also be carried forward by future iGEM teams from our institute, setting a foundation for next year’s project.
| Sample | RBS Sequence |
|---|---|
| PJ102 | CGAAGGAACTACAAAATAAATTATAAGGAGGCACTCA |
| 5 | CTTTAGCAATATACAGGGGGTCCGAAGGAGGTGATTA |
| 6 | TCCTCCCGAATCGCCGCGCTCATCAAGGAGGAGTTTG |
| 7 | CTGCCGCTGCGGACATAAGGCCCTAAGGAGGTGAGCT |
| 8 | GGCGAGGTTTGTCATACCCGCTGAAAGGAGGCCCTCG |
| 9 | TGAGGGAAACGCAATAGAAGTTACAAGGAGGCACTCC |
| 10 | CGAACGAACTACAAAATAATTTCGAAGGAGGCCTACA |
| 11 | ATAAACGGGTTAAAGCGTAAGGTCAGTCCTAATGGAGGTTACAT |
| 12 | TATTAAGATAGGTATCCGACCCCAAGTGCGAATGGAGGCCGTGT |
| 13 | ATATTTTAAAGACTCGCAGACGCAACTTAGAATGGAGGCAGTTA |
| 15 | AGGAAAAAATAATCAAAAAAGATTGTGTCTAAAGGAGGAAAAAA |
| 16 | AGGAGCAAATAATCATAGGAGATTGTGTCTAATCGTAGACAATT |
| 3.1 | CGTCAATTTGGTCGTTTGGTAGTAAAAGGAGGGGCTT |
| 3.2 | GGTGGCCTACCAAAGGCTCTTGATCAAGGAGGTTCCA |
| 3.3 | GAACAGAACATTTCTCGACTTGGAGAAGGAGGTTTAT |
| 3.4 | AGTTTAATCTTCGGCGGTCTTCGATAAGGAGGCCGTT |
| 3.5 | GACTTGACATTGGTCAGAATCAAACAAGGAGGGTATC |
| 3.6 | TACTAGGCAGTCGTTTCGCTAAATAAAGGAGGTGATT |
| 3.7 | CGTCAATTTGGTCGTTTGGTAGTAAAAGGAGGGGCTT |
| 3.9 | AACCCCAGGGCCTAGAGTCGGGGAGAAGGAGGTAGGT |
| 3.10 | GGTGGCCTACCAAAGGCTCTTGATCAACGAGGTTCCA |
| 3.11 | ATTCGAGAAAGTGACGCTCATTCTAAAGGAGGGATCT |
| 3.12 | GGTACTCGTTCTAGATATACTTAGGAAGGAGGCGGAA |
| 3.6 | TACTAGGCAGTCGTTTCGCTAAATAAAGGAGGTGATT |