Hack Yeah Biotech
A second biotech revolution
While technologies in Biotech are proliferating, it becomes easier and easier to keep up. This is not a joke. Admittedly no-one is able is able to keep up with all new inventions, but no-one asks that.
Among all technologies we can choose from, a few have high potential and low barrier to entry. It is a great time to benefit from low cost technology crossovers.
The crossover of biotech, sensors and machine learning holds exceptional promise.
Biotech machine learning plague
I don’t need to convince you of the potential of machine learning in Biotech, and I won’t. On the contrary; I want to warn you.
I believe that most Machine learning projects in biotech are dead before they even start. Many projects never see the light of day because of a lack of data to train models. If there is data, that data is often of insufficient quality to get the thing going.
In most projects there is no business case to justify the efforts needed to obtain the required data. It’s wishful thinking and over optimism that drives these projects.
Crossover optimism
While being pessimistic about machine learning in general, I am optimistic about the crossover of biotech, sensors and machine learning.
The ability to hack sensors and devices creates a tremendous leverage in biotech. It let you create high quality datasets with a fraction of the effort and pursue creative ideas at high pace.
Requirements for machine learning in biotech
Machine learning requires a different approach to devices, sensors, and creativity. In short, different data needs to be generated, measured and stored. Experiments may be repeated more often, and under varying or even dynamic conditions.
In addition you need a high level of control over our devices to support and sustain your team’s creative needs.
Moving instrument suppliers
I have sat in quite some meetings with my client’s instrument suppliers; asking them to adapt their devices to our needs. This is a long process. Sometimes it can be done, but often only to a limited extend and with great efforts. Rename paragraph of positive machine learning Months pass by with demotivating meetings. There is still not a single data point of the kind that we need. This is the start of a painful innovation process.
DIY paradigm
Meanwhile in the area of DIY biotech, 3D printing, laser cutting, and open source software, it gets easier and easier to create, hack and manipulate your own devices.
I recommend you start building or hacking your instruments yourself. It opens a flood gate of ideas and streams of data. Positive energy flows and your team will feel empowered to make real progress.
You will obtain the data that you need to make machine learning work for your product. And your inventions may be patentable and you are likely to find use in other fields.
If you fear the quality of the data that results from some tinkering will not be sufficient, be aware that “quality” in machine learning is often totally different from quality in traditional analytical chemistry or traditional biology. It depends from cases to case, but in general machine learning can cope with more noise and variation than traditional practitioners could ever imagine.
Example: In many cases cheap optical sensors will do as well as expensive once because the noise comes from somewhere else.
Give up on machine learning
If you are not willing to consider changes such as DIY instruments, ask yourself why. If you are reluctant to change or you don’t want to take risk, perhaps you should not start with machine learning at all. Without the required mindset and preparations, machine learning efforts yield negative expected returns. Besides making cost that you will never recover, moral and self esteem will be affected. This negative bias cannot be offset by wishful thinking.
Predictions for the next 5 years
- The crossover of biotech, sensors and machine learning will proliferate.
- AI and machine learning efforts continue to generate negative returns to the unprepared.
- The number of failing machine learning projects in biotech will grow exponentially.
- Biotech companies that apply machine learning successfully will outperform their competitors with large margins.
- In 5 years, companies have learnt not to attempt machine learning projects unprepared.
- Models will stay a commodity; improvements continue to be made quarter on quarter.
- Data is gold, and with the DIY paradigm it becomes profitable to dig it.
- If you dare to adopt a bit of DIY spirit, you will outperform your competitors who stick to traditional high cost, low risk, low reward practices.